An analytical mathematical model for a two-layered catheterized oesophagus is presented in the wave frame. We take due care to conserve the fluids separately. A linear relationship between pressure and flow rate is discovered for the catheterized oesophagus. Pressure and flow rate rise in the presence of a catheter with a thinner peripheral layer. So it can be suggested that no patient should be fed anything directly through the mouth once a catheter has been inserted into the oesophagus.

Phenotypic heterogeneity of melanoma cells contributes to drug tolerance, increased metastasis, and immune evasion in patients with progressive disease. Diverse mechanisms have been individually reported to shape extensive intra- and inter- tumoral phenotypic heterogeneity, such as IFNγ signaling and proliferative-invasive transition, but how their crosstalk impacts tumor progression remains largely elusive. Here, using dynamical systems modeling and analysis of publicly available transcriptomic data at both bulk and single-cell levels, we demonstrate that the emergent dynamics of a regulatory network comprising MITF, SOX10, SOX9, JUN and ZEB1 can recapitulate experimental observations about the co-existence of diverse phenotypes (proliferative, neural crest-like, invasive) and reversible cell-state transitions among them. These phenotypes have varied levels of PD-L1, thus driving variability in immunosuppression. We elucidate how this heterogeneity in PD-L1 levels can be aggravated by combinatorial dynamics of these regulators with IFNγ signaling. Our model predictions are corroborated by analysis of PD-L1 levels in pre- and post-treatment scenarios both in vitro and in vivo. Our calibrated dynamical model offers a platform to test combinatorial therapies and provide rational avenues for clinical management of metastatic melanoma.

Cancer metastasis accounts for nearly 90% of cancer-related deaths. Epithelial to Mesenchymal Transition (EMT) of cancer cells is key to cancer metastasis. Recent preclinical and clinical data suggest that hybrid E/M cell states harbor the highest metastatic fitness instead of the fully mesenchymal cells. However, the mechanistic details of their survival strategies during metastasis remain unclear. We model the dynamics of a minimalistic gene regulatory network (GRN) between regulators of EMT and PD-L1, a known immune suppressor. We show that hybrid E/M states are highly likely to exhibit high PD-L1 levels, like those seen in mesenchymal cells, thus obviating the need to undergo a complete EMT to cause immunosuppression. We show that the switch from an epithelial/low-PDL1 state to a hybrid/mesenchymal state with high-PDL1 is reversible. We demonstrate that acquiring phenotypic resistance to targeted therapy can co-occur with high PD-L1 levels, enabling cross-resistance and enhancing breast cancer cell fitness during metastasis. We validate our model predictions by extensive analysis of transcriptomic datasets across multiple cancers at bulk and single-cell levels. Our results highlight how the emergent dynamics of interconnected GRNs can coordinate various axes of cellular fitness during metastasis, thus laying the foundation for the rational design of clinical therapies.

For a better understanding of the metabolism of an organism, it is crucial to build detailed mathematical models. The availability of omics data in the past decade helped to improve our understanding of metabolism through genome-scale metabolic models (GEMs). To capture the reactions that are active in a given condition, transcriptomics are integrated into GEMs to build context-specific models (CSMs). A context here could refer to any perturbation that can alter the gene expression levels. Based on the expression levels of the genes and the Gene-Protein-Reaction rules, the core reactions that are known to be active in the given context are identified. However, noisy data, improper thresholding, and lack of genetic evidence for spontaneous and diffusion reactions often result in an incomplete draft of a CSM that has only the core reactions. In this study, we developed three distinct algorithms to build and analyse the CSMs from GEMs in a rapid manner. The first algorithm, SprintCore, integrates transcriptomics into GSM to construct CSM. The second algorithm, SprintCC, checks for consistency and reports blocked reactions in a metabolic reaction network, and the third algorithm, SprintTag, tags all the reversible reactions as reversible or pseudo-irreversible. SprintFamily of algorithms outperforms the previous algorithms.

The bacteria of the Bifidobacterium genus are commonly used as probiotics, that is, live microorganisms that confer a health benefit on the host, although the specific functional effects depend on each strain. It is known that the functionality of some probiotic bacteria is related to the presence of some effector molecules, but the detailed description of the molecular mechanism of action and the genetic basis of its synthesis has only been described in a few cases. In this work we analyze the commercial probiotic B. animalis subsp. lactis BPL1 (CECT 8145), which reduces fat content and triglycerides and modulates oxidative stress and feeding behavior. Here we propose an in silico approach based on genome-scale metabolic modeling to identify and describe the metabolic pathways, metabolites and genes of interest involved in the mechanism of action of BPL1. In addition, we have performed a transcriptomic study along different growth phases of this strain in both functional and non-functional conditions, that reveals the possible genes involved in its antioxidant properties. With all of this, the aim is to integrate these data in the metabolic model to better understand its metabolism and to improve the production of the effector compounds responsible for the functionality

Given the close association of Type 2 diabetes (T2D) with Alzheimer’s disease (AD), elucidating the molecular and epigenetic regulatory mechanisms that trigger the progression of T2D towards AD is a dire need. However, the knowledge of regulatory processes is scattered. In addition, predictive models to project the progression of T2D to AD are unavailable.
We curated the genes, transcription factors (TF) and microRNAs (miRNA) associated with T2D and AD from various databases. The significant regulatory pairs were analyzed using cumulative hypergeometric test to generate a miRNA-TF-gene regulatory feed-forward loop (FFL) network. Differential DNA methylation, differential gene expression, alternate splicing and TF phosphorylation features extracted from multiple datasets of AD and T2D were incorporated to generate deterministic model. Further, this model was used to simulate the switching of T2D to AD.
2 TFs and 11 miRNAs exhibited FFL regulation of 29 genes in both AD and T2D. In T2D, TFs regulation was more pronounced while in AD, miRNA regulation was prominent. Likewise, differences in methylation, splicing and TF phosphorylation were observed. The simulation of deterministic model generated using these parameters demonstrated the progress of T2D towards AD. The model framework could be used to assess the theoretical treatment plan.

Genome-scale metabolic networks (GSMNs) allow metabolic modelling of living cells. Additionally, methods have been designed to create context-specific GSMNs representing specific tissues or cell types. These methods use optimization algorithms to identify a subnetwork which most closely matches experimental-omics data. However, these algorithms generally do not produce enough constraints to generate one unique optimal subnetwork, but instead result in a set of optimal solutions. Ignoring this variability can lead to a loss of information and alter the biological interpretation of downstream analysis. In such cases, a set of optimal subnetworks can be enumerated to obtain more complete information about the metabolic state.
Previously, Rodríguez-Mier et al. introduced DEXOM as a unified approach for enumeration of context-specific GSMNs aiming to obtain a diverse subset of the optimal solutions. Several enumeration methods were developed and compared regarding their coverage of the optimal solution space and their applicability for biological interpretation.
OCMMED is a new DEXOM-based workflow for generating cell-specific GSMNs, assembled in snakemake for better reproducibility and scalability with computer clusters. After inputting experimental transcriptomics data from one cell type, subnetworks are enumerated before being merged into one singular GSMN which represents the full potential metabolic variability of the studied cell type.

Hybrid system modelling framework can be regarded as a natural choice for description of gene regulatory networks since it can provide a seamless integration of both discrete and continuous aspects of biological system’s behaviour. In this work we explore the potential of hybrid modelling framework to identify conditions that irrevocably leads to different dynamical behaviours of the system. Such behavioural regions can be interpreted either as evolutionary differentiation between cell types, or transition of the modelled system into one of several known distinct stable biological processes. Our previously developed models for phage viruses allow to identify transitions and genes that are triggering them, which irrevocably leads the viruses either to lytic or lysogenic behaviours (the triggering factors are model specific and differ for lambdoid and Mu phages). A newly developed hybrid model for haematopoietic cell differentiation focuses on cell proliferation stages from common myeloid progenitor and unambiguously identifies triggering factors behind erythrocyte/megakaryocyte and granulocyte/monocyte differentiations. An unanticipated result is the observation that, although the pathways leading to differentiation to intermediate progenitors are well defined, the activity of the same genes can play different roles depending on the stage at which their activity become a determining evolutionary factor.

Computational modelling of cell metabolism is typically based on representing metabolism as a network and using constraint-based modelling to identify reactions that exhibit significantly different usage between biological samples. Many methodologies have been proposed to integrate post-genomic data into these models, but few have explored the integration of single-cell RNA sequencing (scRNA-seq) data. This is challenging because scRNA-seq data are more subject to noise and false zero values, and the definition of a proper objective function is even more critical when dealing with heterogeneous single cells.
To address these challenges, we propose a corner-based sampling method to sample the feasible region of a metabolic network customized with scRNA-seq data. This method is designed to avoid the impact of under-sampling, which can lead to statistically significant differences even when analyzing samples from the same feasible region. Additionally, we highlight the importance of denoising read counts for more reliable predictions.
This talk summarizes the best practices for sampling a metabolic network customized with scRNA-seq data. We emphasize the importance of proper sampling to compare and characterize metabolic heterogeneity across different samples. By avoiding under-sampling and denoising read counts, we can generate more reliable predictions and better understand metabolic heterogeneity in single cells.

In this work, we blueprint a Dashboard that allows users to simulate the bacterial community’s evolution through an intuitive GUI. The underlying Python-coded simulator implements an agent-based model of bacterial species, nutrients, and environment, allowing full customization and upgradability of the tool, due to its intrinsic modularity. Specifically, the model aims to represent discretized spaces, hosting a certain number of bacteria for each species and a defined amount of nutrients characterizing the surrounding environment. Bacteria can migrate from one spatial unit into another, looking for different nutrients (i.e., metabolites) across the whole space path. Growth and survival are governed by bacterial metabolisms, which are in turn functions of the metabolites present in each specific spatial unit at a certain time. Thus, our tool simulates how bacteria consume and produce metabolites, following species-specific metabolism rules, letting the system dynamically evolve through bacterial growth, death, spatial migration, and continuous updates of the available metabolite pool.

Collecting time-course data in vivo to study the dynamics of whole-body metabolism and gene expression can be ethically questionable and technically impractical. To explore the potential metabolic pathways involved in inter-organ communication and systemic metabolic homeostasis in the context of the circadian cycle, we have developed a lightweight framework based on Flux Balance Analysis and Metabolic Networks.
We utilized a modified version of the HGEM1 metabolic network as a scaffold and developed a new dynamic FBA algorithm that incorporates information on systemic metabolic concentrations and thermodynamics. Our optimization process considers critical phenomena like osmotic balance and allosteric interactions to maintain system homeostasis. To generate tissue-specific networks, we integrated data from multiple sources and obtained good approximations for the liver, skeletal muscle, and white adipose tissues. We also utilized published multi-omics datasets using mouse tissues to further refine and fine-tune the model parameters with a genetic algorithm.
Our goal is to provide a more in-depth study of inter-organ communication, considering also unperturbed states with a focus on dysfunctional circadian clocks and dietary challenges. By bridging the gap between sparse time points collected from in vivo samples, our method has the potential to elucidate the mechanisms underlying systemic metabolic regulation.

Cancer progression is shaped by the interplay between tumor and surrounding host cells.
For instance, cancers cells might stimulate angiogenesis, supporting tumor growth, while immune cells may infiltrate the tumor, slowing its spread. Thus, it is essential to better understand such cellular interactions in the context of the surrounding tissue, known as the tumor microenvironment (TME). Modern proteomic imaging techniques provide a map of cellular heterogeneity and spatial organization of the TME. Thanks to Multiplexed Ion Beam Imaging (MIBI), we are able to profile simultaneously the abundance and spatial distribution of up to 40 proteins in single cells from intact tissue samples. Our lab established MIBI protocols to infer cell identity and quantify the expression of key regulators in multiple metabolic pathways across all cells in multiple tissues. Here, we explore computational approaches to best represent shared cellular metabolic programmes and interactions in the TME, which we find to cluster in local niches. Further, we aim to explain the molecular origin and consequences of such patterns, and to link these findings to disease stratification and therapeutic responses.

The immune system’s role is to efficiently recognize and eliminate foreign agents. This is achieved through various mechanisms, including the labeling of pathogens with opsonins, as well as the secretion of antimicrobial peptides (AMPs) by immune cells that can kill pathogens extracellularly. However, some pathogens, such as the yeast Candida albicans, can evade the immune response.
This study proposes a modeling approach to investigate pathogens’ potential immune evasion mechanism, called spatial distancing, where the pathogen secretes molecules that can bind to AMPs, forming complexes that diffuse away from the cell. This model suggests that microbial pathogens can evade the immune system based on molecular complex formation and concentration equilibration by complex diffusion. We applied and compared two modeling approaches to represent the reaction and diffusion of molecules, (i) with partial differential equations, and (ii) within an agent-based model.
The time-dependent spatial distributions of the various molecules revealed that the secretion of molecules by the pathogenic cell induces indeed a reduction in the concentration of AMPs in the close vicinity of the microbial cell. This phenomenon was observed across a wide range of parameter values, suggesting that spatial distancing could be a robust and effective immune evasion mechanism pathogens use.

Precision medicine requires High-Performance Computing (HPC) platforms to model complex and massive volumes of biomedical data. In particular, multiscale cell simulators have proven useful thanks to their ability to help uncover and explain disease mechanisms.
PhysiCell is an open-source cell simulator of many interacting cells that respond to and influence their microenvironment. The simulation size scale of these tools is an open problem whose solution is attached to the efficient usage of HPC resources. While the state of the art has reached the tissue level, i.e., simulations up to 109 cells, the ultimate goal in this area is represented by larger and more realistic simulations collectively called Digital Twins.
We have built a hybrid OpenMP-MPI implementation of PhysiCell to distribute the simulation across multiple computation nodes and named it PhysiCell-X. The distribution of the domain and its optimisation allows larger simulations to be modelled across compute nodes, potentially enabling real-sized tumour simulations.
Modelling is among the most promising techniques in precision medicine, but it is also a very demanding task in terms of energy resources. Thus, the field would benefit from examples of tools properly adapted and optimised to HPC platforms enabling the realisation of Digital Twins.

Genomic heterogeneity is a distinctive feature of cancer diseases, affecting the efficacy of medical treatments and leading patients to relapse. Tumourigenesis emerges as a strongly stochastic process, producing a variable landscape of genomic configurations that build the global identity of tumours. In this context, Matrix factorisation techniques represent a suitable approach to understanding such complex patterns of variability, identifying latent patterns that represent the basic building blocks of provided observations. To this end, we develop a hierarchical factorisation model conceived from a systems biology perspective, which integrates the topology of signalling pathways. Our model simultaneously decomposes gene and signalling pathways activities, revealing the molecular strategies used by individual tumours to develop the hallmarks of cancer. We applied our protocol to a cohort of breast cancer patients recapitulating the internal composition of some of the most relevant altered biological processes in the disease, such as the epidermal growth factor deregulation in the Her2 subtype or the differences between the Luminal A and Luminal B subtypes in oestrogen response and the cell cycle regulation. We envision hierarchical matrix factorisation designs will be essential to understand how phenotypically similar cancers arise from very different genomic configurations.

Single-cell sequencing has revolutionised the way molecular biology research is conducted, providing comprehensive portraits of complex and dynamic cellular processes, such as cell differentiation or tumour growth. Computational analysis of single-cell data is becoming standard thanks to the definition of a set of best practices and the development of popular frameworks such as Seurat or Scanpy. Furthermore, single-cell studies have contributed to pioneering the “cell atlas” concept, describing the cellular diversity of human tissues. Although these approaches have provided important contributions to clarifying the essential cellular processes behind human diseases, there is still a growing interest in tools that approach the analysis of cellular variability from a Systems Biology point of view, understanding biological systems as a whole, rather than focusing on their constituent elements. With this perspective, we developed Sybarite (https://github.com/jcarbonell-bsc/sybarite), an R package that integrates different modelling approaches to quantify the activity of molecular pathways and gene regulatory networks. Sybarite was tested with different scRNA-Seq datasets finding systemic patterns that exhibit differential activity across cell types. Finally, Sybarite established mechanistic connections between the different modelling approaches, hence providing a more holistic description of cell functioning alterations

Metastasis is defined as the spread of cancer cells from a primary tumor to a distant site in a patient’s body. It is a major cause of mortality for patients suffering from solid cancers. Yet, the evolutionary dynamics driving metastasis formation and the timing of successful spreading are poorly understood.
We introduce MetMHN, a continuous time Markov chain model that describes the sequential accumulation of aberration events, such as mutations and copy number changes, in matched primary tumor and metastasis samples. It is an extension of the Mutual Hazard Network (MHN) model, which describes primary tumors only. MetMHN additionally models the temporal evolution of a matched metastasis, which shares its initial history with the primary tumor and then evolves independently after an (unobserved) seeding event. We fit the model on cross-sectional matched pancreatic cancer samples from the MSK-Impact dataset (Nguyen et al.). With this model we answered questions such as: What’s the likeliest order of mutation occurrences in an observed sample? How likely is it that a certain mutation happened prior to metastatic spread? Given an observed primary tumor sequence, how likely is it that an (unobserved) metastasis is present?

Genome-scale models of metabolism are powerful tools for the analysis and contextualisation of omics data in microorganisms. However, although most environmental and human-related microbes are unculturable and difficult to isolate, modelling approaches for omics data integration are limited to individual microbes and simple artificial co-cultures. Here, we present an approach that extends condition-specific metabolic modelling to microbial communities by metatranscriptomics data incorporation in metagenome-scale models. The approach is used as a core step of a culture-independent pipeline for modelling microbial activity and metabolite exchanges based on the reconstruction of metagenome-assembled genomes and associated genome-centric metatranscriptomes. We target and validate this pipeline on the study of anaerobic digestion consortia driven by hydrogen availability and human gut microbiota dysbiosis associated with Crohn’s disease. In both scenarios, metatranscriptomic data integration allows more accurately estimating microbial growth rates and the global production of main metabolites. Further, our models identify condition-dependent amino acid requirements in archaeal species and a reduced short-chain fatty acid exchange network associated with disease. The newly developed approach can be applied in culture-independent settings, starting to disclose multi-omics metagenome-scale modelling and providing a new tool for investigating natural microbial ecosystems.

Flux sampling is a powerful technique for analyzing the feasible spaces of constraint-based metabolic models. We recently implemented CuFluxSampler.jl (https://github.com/LCSB-BioCore/CuFluxSampler.jl), a collection of GPU-accelerated flux sampling algorithms that aid quick exploration of complex constraint-based metabolic models.
In the poster, we highlight the existing uses of CuFluxSampler.jl, and detail a newly developed extension that allows to sample near-optimal spaces and density distributions from metabolic modeling tasks such as parsimonious flux balance analysis and minimization of metabolic adjustment, which require handling of quadratic objectives and bounds.
CuFluxSampler.jl is built upon COBREXA.jl and utilizes its main design features, making it easy to run massively parallel analysis of models subjected to diverse conditions and constraints. We further demonstrate this by exploring ensembles of metabolic flux samples from models that are constrained by parsimonious enzyme allocation, showing the ability to also sample directly from the enzyme amounts and other phenotypic properties.

Many infectious diseases, including SARS-COVID19, Zika, Hepatitis C, and Poliomyelitis, are caused by positive-strand RNA (+RNA) viruses. Despite their different strategies, +RNA viruses share common life cycle processes within the host cell. A unified and quantitative analysis of these viruses can identify common bottlenecks.
In this study, we develop a generalized dynamical model to monitor viral molecule levels within the cell and estimate lifecycle determinants for several +RNA viruses. The model considers the effects of viral mutations, drugs, and host cell permissivity on viral lifecycle dynamics.
Stochastic simulations of the model demonstrate that if seeding viral RNA degrades before establishing a robust replication machinery, infection can be extinguished. We also evaluate the effect of virus-host processes and viral seeding on cellular infectivity. For example, cytoplasmic RNA degradation and intracellular rearrangements that facilitate viral replication reduce cellular infectivity, especially when combined. Our model predicts that synergy among these parameters limits +RNA virus infection and suggests new avenues for inhibiting infections by targeting the early lifecycle bottlenecks.

Bacteria of the Salmonella genus are intracellular pathogens, which cause gastroenteritis and typhoid fever in animals and humans, and are responsible for millions of infections and thousands of deaths across the world every year. Furthermore, Salmonella has played the role of a model organism for studying host-pathogen interactions. Within epithelial cells, there are two distinct subpopulations of Salmonella: (i) a large fraction of Salmonella, which are enclosed by vacuoles, and (ii) a small fraction of hyper-replicating cytosolic Salmonella. Here, by considering the infection of epithelial cells by Salmonella as a discrete-state, continuous-time Markov process, we propose a stochastic model of infection, which includes the invasion of Salmonella into the epithelial cells by a cooperative strategy, the replication inside the Salmonella-containing vacuole, and the bacterial proliferation in the cytosol. The xenophagic degradation of cytosolic bacteria is considered, too. The stochastic approach provides important insights into stochastic variation and heterogeneity of the vacuolar and cytosolic Salmonella populations on a single-cell level over time. Specifically, we predict the percentage of infected human epithelial cells depending on the incubation time and the multiplicity of infection, and the bacterial load of the infected cells at different post-infection times.

My PhD project aims to address the challenge of personalized medicine for cancer therapeutics through the development of an innovative approach combining bioinformatics and mathematical modeling. The cancer community agrees on the need for patient-tailored therapy, which requires the design of a digital representation of the patient including tumor omics or treatment history. Such approach is being developed in the context of deficiencies of the SWI/SNF epigenetic complex, which appear in 20% of all solid tumors, highlighting its pivotal role in tumorigenesis and making it a potential therapeutic target. Little is known about how to selectively target defects in this complex, so it is crucial to unravel genetic vulnerabilities associated to SWI/SNF deficiencies.
Firstly, multi-omics, drug screening and CRISPR data available in several SWI/SNF deficient cell lines have been analyzed with a newly developed enrichment pipeline to identify the most deregulated pathways in a given tumor. Then, ordinary differential equations mechanistic models are being built and calibrated to experimental data to represent those pathways in a dynamical manner and predict optimal drug combinations. Finally, optimal drug combinations will be tested experimentally to validate their efficacy and safety and the approach will ultimately be translated into the clinics using patient data.

To help map the field of agent-based modelling for digital twins, at PerMedCoE we gathered different developer teams and organised a community-driven benchmark. The tools that participate in this benchmark were PhysiCell (Ghaffarizadeh et al., 2018), Chaste (Cooper et al., 2020), BioDynaMo (Breitwieser et al., 2021) and TiSim/CellSys (Hoehme and Drasdo, 2010).
The goal of the benchmark was to agree on a set of reference datasets, metrics and scope of the scientific questions addressed by the tests and run these in all the tools in a common computing cluster.
Even the simple unit tests yielded different results among the tools, but the tools fitted well a set of experimental growth values of a 2D monolayer growing in vitro.
From the results of these, it was decided that the next steps were to carefully study the simulation results of each tool, their code implementation and their underlying mathematics to be sure that the benchmark is comparing tools that simulate exactly the same behaviour using the same equations.
These outcomes will be disseminated in a community paper with a global picture of where we stand, identifying gaps and obstacles that need addressing if we are to deliver digital twins in the future.

Although the clinical efficacy of Ibrutinib in Chronic Lymphocytic Leukemia (CLL) partly rests on metabolic alterations (PMID:32581549), a systems-level understanding of these changes in CLL and surrounding cells is lacking. To address this, we analyzed publicly available scRNA-seq data from CLL blood samples (Pre-Ibr., Post-Ibr 30 days; n=4 cases) (PMID:31996669) by designing a new pipeline using Seurat, SingleR cell-annotation tool, scFEA neural-networks (cell-wise fluxomics) (PMID:34301623) and MetaboAnalyst enrichment platform. scFEA converted the typical Genes X Cells matrices to Metabolic Modules x Cells matrices. Through PCA, UMAP, differential module fluxomics, and metabolite enrichment on the latter matrices, we detected suppression of TCA cycle and increases in glycolysis/Warburg effect and pyrimidine synthesis in post-Ibrutinib CLL B cells. Increased glycolysis was also dominant in post-treatment CD4/CD8 T cells, hematopoietic stem cells, megakaryocytes, and myeloid progenitors. Interestingly, enhanced purine synthesis and import of oxoglutarate/malate were enhanced in post-Ibrutinib CLL B and T cells. These preliminary in-silico findings point towards – insofar nascent – metabolic adaptations of CLL and adjacent cells post-Ibrutinib. The theranostic value of these early data merits further investigation considering novel metabolic biomarkers and the use of metabolic modulators in CLL cases with suboptimal Ibrutinib responses.

Myocardial infarction (MI) is a severe threat worldwide, characterized by a coronary artery obstruction disrupting heart perfusions. The survival of the patient is determined by the blood flow re-establishment. Paradoxically this latter induces cell death and increasing inflammation. Progress in clinical care has reduced cell death, whereas inflammation is thought to participate in secondary events such as heart failure. Therefore, a better understanding of the early mechanism of inflammation is required.
Our goal is to characterize the temporality of gene transcription during ischemia-reperfusion, to further identify candidate targets for therapeutics.
A RNAseq cohort has been designed with 8 mouse hearts samples at 12 early times, from the first 5 minutes of ischemia to 24 hours after reperfusion.
7,500 transcripts were identified to be differentially expressed over time. DETs were clustered by similarity of their temporal expression by time-series clustering. An heatmap reports the timely cascade of transcripts responses. To characterize each identified kinetic, we created reduced time profiles with significant breakpoints.
To run this work at the next level, we will build networks with functional annotations and model dynamic subnetworks of transcription within functional annotations. This will lead to a comprehensible illustration of transcriptional kinetics during ischemia and reperfusion phases.

The human immune system constantly has to fight microbial invaders, such as the pathogenic fungus Aspergillus fumigatus. Humans inhale hundreds of conidia daily, which can reach the lower respiratory tract. If not efficiently cleared, they form hyphae within hours, resulting in life-threatening infections like invasive aspergillosis.
We previously developed a computational hybrid agent-based model (hABM) to simulate virtual infection scenarios of A. fumigatus in the lung. The hABM provides a realistic, to-scale representation of one alveolus, consisting of a ¾ sphere, pores of Kohn, alveolar epithelial cells (AEC), and alveolar macrophages (AM). The model includes conidia-induced chemokine secretion by AECs, which is sensed by AMs, directing their migration towards the infection.
The current study is extending our most recent results, in which we utilized the hABM to examine the number of AMs in the lung. We extended the hABM to capture the phagocytic activity of AECs and hyphal growth of germinating conidia, which provides a more in-depth representation of host-pathogen interactions. Applying our hABM in thousands of virtual experiments, we are investigating the individual contributions of AM and AECs in clearance of A. fumigatus infections. This contributes to uncovering the defense mechanisms of the early immune response in the lung.

Photorhabdus is a Gram-negative bacterial genus containing both potent insect and emerging human pathogens. Most insect-restricted species display temperature restriction, unable to grow above 34°C, whilst Photorhabdus asymbiotica, an emerging human pathogen causing Photorhabdosis, can grow at both 28°C and 37°C to infect insect and mammalian hosts, respectively. A metabolic shift has been proposed to facilitate survival of this pathogen at higher temperatures, yet the biological mechanisms and processes underlying this shift are poorly understood. This study has reconstructed the genome-scale metabolic model of P. asymbiotica (iEC1073). iEC1073 is an extensively manually curated metabolic reconstruction, validated through in silico gene-knockout and nutrient utilisation experiments. Integration of iEC1073 with transcriptomics data obtained for P. asymbiotica at temperatures of 28°C and 37°C allowed the development of temperature-dependent reconstructions. These networks represent the metabolic adaptations the pathogen undergoes when shifting to a higher temperature in a mammalian compared to insect host. P. asymbiotica potentially undergoes a stringent response to the stress induced by nutrient deprivation in the mammalian host, characterised by the upregulation of the purine metabolism pathway including reactions synthesising the precursors for the signalling molecule guanosine penta/tetradiphosphate which modulates transcriptional changes associated with the stress response and virulence.

MaBoSS is a software for simulating signaling and regulatory networks, which produces trajectories describing the evolution of the states’ probabilities of a Boolean model. We present a GPU-accelerated version of MaBoSS that provides a speed-up of over 200 times for the evaluation of the Boolean network transitions. As the main bottleneck, aggregation of the simulation statistics becomes a major factor in the algorithm run-time that would deny further parallelization. We demonstrate several ways how the speed-up substantiates novel aggregation algorithm with different modes of exploring the trajectories and fixed points of the Boolean network simulations. We detail an example where fast dynamic aggregation of the Boolean network states into buckets in each time window (similar to online K-means clustering) removes most of the aggregation overhead, and the performance improvement is sufficient to enable the users to interactively expand and explore the result statistics even with very large models. We hope this will aid exploration of the complex space of perturbed models (esp. multiple mutants).

Grouping gene expression into gene sets representing biological functions provides better insights than studying individual genes. Existing approaches to project gene expression into gene set scores cannot define scores with a consistent definition across different datasets and technologies.
We present NetActivity, a machine learning framework that generates gene set activity scores (GSAS) based on a sparsely-connected autoencoder, where each neuron of the inner layer represents a gene set, and a specialized three-tier training. We considered 1,518 GO biological processes terms and KEGG pathways to define the inner-layer of NetActivity, and trained it using all GTEx samples.
NetActivity generated GSAS robust to the initialization parameters, representative of the original transcriptome, and gave higher importance to more biologically relevant genes. Moreover, compared to GSVA and hipathia, state-of-the-art methods, NetActivity returned GSAS with a more consistent definition. Finally, NetActivity enabled combining bulk RNA-seq and microarray datasets in a prostate cancer progression meta-analysis, highlighting gene sets related to cell division, key for disease progression. When applying NetActivity to metastatic prostate cancer, samples resistant to abiraterone treatment presented GSAS differences in the same gene sets identified in the prostate cancer meta-analysis, while a classical gene set enrichment analysis identified gene sets uninformative for prostate cancer.

Cancer progression is a complex phenomenon that spans multiple scales from molecular to cellular and intercellular. Understanding cancer biology and the emergence of resistance can benefit from using multi-scale models which enable studying the interplay between molecular processes, population dynamics, and the microenvironment. Herein we introduce PhysiBoSS-2.0, a simulation addon that expands the PhysiCell multi-scale modelling framework functionalities by allowing the integration of cell signalling and regulatory network models within the cell agents. Using PhysiBoSS-2.0 we implemented a model of the gastric adenocarcinoma cell line AGS for simulating in-silico drug screening experiments and investigating resistance mechanisms. Specifically, we integrated a published Boolean model of AGS together with experimental data including, cell line-specific doubling time, basal apoptotic rate, as well as gene expression. Unknown parameters were calibrated by fitting simulations to experimental growth curves. We then simulated the growth of the AGS cell line treated with different drugs and found that our results quantitatively reproduce the experimentally measured time course. Finally, we extended the model with an efflux pump-mediated resistance mechanism and used it to explore different therapy strategies. Altogether, our results show how PhysiBoSS-2.0 can be used to simulate realistic drug screening experiments.

Understanding the microbiome, an indispensable part of the human and environment will help to manipulate and manage the microbiome for their application in health care and sustainable systems. Community modelling uses genome-scale metabolic models (GSMMs) to infer the communications between the microbial species in a microbiome. However, in the absence of species-resolved metagenomic data from 16S rRNA sequencing studies, our ability to perform metabolic predictions and inferences is limited. To address this gap, a novel methodology called PAGER (PAn-GEnome Reconstruction) is developed to construct a genera-specific metabolic model to explore the metabolic potential of the genus and gain deeper insights into the community structure-function relationship. The reconstructed PGMM is validated by assessing their ability to retain the metabolic capabilities of an individual species and represent the functionalities of the microbial genus in a community. The analysis exhibits that the flexible nature of PGMM expands our knowledge of the metabolic niches of a genus, allowing for investigation of the genus-metabolic landscape. PGMM can be employed with GSMM for building microbial communities to reduce uncertainties and improve the prediction accuracy of metabolic interactions in microbiota.

Microbes, both as monocultures and communities, are increasingly being used to produce valuable chemicals. While monocultures have been more successful, microbial communities offer advantages such as robustness and reduced metabolic burden. However, they are complex and challenging to control and manipulate. Therefore, there is a dire need for systematic, computational approaches to design bioprocesses with communities. We have developed a dynamic Flux Balance Analysis (dFBA) based algorithm to study and compare the biosynthetic capability of communities and monocultures. This enables us to choose the best system based on the difference in biosynthetic capabilities. Simulations of 80 two-member communities and their monocultures have identified communities that can produce products that are either not secreted or secreted in low levels by the monocultures.
Rational strain design of communities can further improve bioprocesses. However, existing computational algorithms cannot be directly used to metabolically engineer communities. We have developed an algorithm to identify deletion and amplification strategies for communities. We are also working on developing mutually auxotrophic communities and studying co-production in communities using similar approaches. Together the suite of algorithms aids us in designing better biological systems for bioprocessing.

Reactive Species (RS) like hydroxyl radical, superoxide anion, nitric oxide radical, etc are important regulatory molecules that are highly reactive in nature. Their roles get pronounced under oxidative stress conditions. Constraint-based models using Genome-Scale Metabolic (GSM) models are used to predict phenotypes of microbes, tissues, cancer cell lines etc. Despite the prevalent roles of RS, their contribution to GSM models of humans and microbes is very limited. We have developed a scalable module of RS reactions relevant to human metabolism and have built RS module integrated GSM models of cancer cell lines and human-macrophage model. Metabolic rewiring is a hallmark of cancer and causes redox imbalance. Thus, RS levels are altered during tumour development. The RS integrated cancer GSM models have outperformed their GSM model counterparts in the prediction of cancer phenotypes. The regulation of ferroptosis, a recently recognised form of regulated cell death, that is characterised by the accumulation of iron and lipid peroxides was better highlighted in the RS integrated cancer cell line GSM models. RS-integrated macrophage model with bacteria to study host-pathogen interaction had altered fluxes through importable metabolic pathways like fatty acid metabolism, glycerophospholipid metabolism, lipopolyaccharide metabolism etc, which can be targeted for pathogen clearance.

The internalization of two endodermal precursor cells during early gastrulation in C. elegans provides a simplified context to study the physical mechanism of cellular ingression in detail. While apical constriction has been recognized as the primary mechanical driver, the potential contributions of other mechanisms, such as force generation in the covering cells and coordinated cell divisions, have been largely overlooked.
Our study combines a large number of full embryo 3D cell segmentations with a cellular force model, allowing us to perform mechanical simulations. By simulating gastrulation under various scenarios and comparing the results to measured cell shapes, we can test the apical constriction mechanism and explore additional hypotheses, such as whether concurrent cell divisions facilitate ingression. To validate these hypotheses, we predict the effect of perturbations and compare it to experimental observations, for example via knockdown of cellular adhesion molecules. Additionally, we enhance our analysis by a comprehensive characterization of cell shapes, cellular movements and cortical protein concentrations, including myosin and cadherin to ultimately provide a detailed perspective on this archetypical example of cellular ingression.

Background: Microchloropsis gaditana is a promising microalga for biofuel applications due to its ability to accumulate a high level of lipids. In this work, a new genome-scale metabolic model of M. gaditana was reconstructed, manually curated, and validated.
Results: The model, iMgadit884, encompasses 884 genes associated with 2,324 reactions and 1,973 metabolites distributed across eight compartments: extracellular, cytosol, chloroplasts stroma and lumen, endoplasmic reticulum, peroxisome and mitochondrial matrix and intermembrane space. Membrane and storage glycerolipid biosynthesis and degradation pathways were exhaustively described and account for 43.5 % of model reactions. Based on iMgadit884 content, two-dimensional pathway maps were drawn, providing a systems-level visualization of M. gaditana metabolism.
iMgadit884 was effective in capturing M. gaditana growth in various conditions with good agreement between experimental and predicted data. Biomass composition of two M. gaditana strains were added to iMgadit884: a WT strain and a strain harboring a point mutation in MgACSBG (Naga_100014g59) gene, leading to a significant variation of fatty acid and glycerolipid profiles. In silico flux distributions using each biomass reaction as objective function were predicted and compared to further analyze mutant phenotype.
Conclusion: iMgadit884 constitutes a powerful tool for further characterization of M. gaditana metabolism and for model-driven strain design.

Tightly controlled changes in cell shape underlie cellular motion and self-organization in processes as diverse as wound healing and embryogenesis. Cell shape arises through contractility of the actomyosin cortex, interactions with the environment like adhesion, and active dynamic processes like cell division and protrusions. Quantifying these forces is a major challenge.
We propose an innovative approach to infer cellular forces from cell shape. We start with confocal fluorescence microscopy time-lapses of C. elegans embryos. After segmentation, the cell shapes are introduced into a numerical simulation that employs a biophysical model of cell shape. We then optimize the system by running simulations until a force landscape is found that explains the cell shapes. To experimentally validate our inferences, we performed cortical laser ablation experiments on early embryonic cells.
By applying this method, we could construct a timeline of force generation based on many embryos without invasive experimental measurements. This pipeline facilitates generating large amounts of data to analyze morphogenesis, the cellular effects of gene knockouts and to associate protein localization with force generation.

The eukaryotic cell cycle is driven by waves of cyclin-dependent kinase (cyclin/Cdk) activities that rise and fall with a timely pattern called “waves of cyclins”. This pattern guarantees coordination of DNA synthesis with cell division. Through computational modelling, we have recently identified a minimal network underlying cyclin/Cdk1 autonomous oscillations in budding yeast. Here, we first explore whether these oscillations may be achieved by synthesizing a functional genome consisting of a minimal set of cell cycle genes: G1 cyclins, mitotic cyclins, and their positive and negative regulators. Selected genes are pairwise deleted by CRISPR from their native loci in the yeast genome and simultaneously relocated with their native promoters into a synthetic gene cluster in the same cell. We then remove combinations of genes from this cluster, and their frequency of gene loss and growth rates are compared to kinetic models of the cyclin/Cdk1 network verified against quantitative data of cyclin dynamics. We unravel a novel molecular design that synchronizes Clb/Cdk1 oscillations. Through integration of Synthetic and Systems Biology, this work shows that the genetic complexity of the yeast cell cycle can be reduced to reproduce cell cycle oscillations, identifying a novel molecular network underlying cell proliferation dynamics.

Introduction: Understanding the impact of SARS-COV-2 evolution on T cell evasion in an HLA-dependant manner is crucial to identify human subgroups at risk of a reduced T cell response to SARS-CoV-2 variants. We hypothesize that the HLA diversity in the human population leads to variations in T cell effectiveness against existing and emerging SARS-COV-2 Variants of Concerns.
Methods/results: We developed a computational framework to track the diversification of SARS-CoV-2 T cell epitopes while assessing the impact of emerging variants on CD8+ epitope presentation, taking in consideration all 6 HLA class I alleles from a wide range of HLA-typed individuals from two biobanks (RECOVER-2, n = 611; UK-Biobank, n = 487,000). Preliminary data indicates that T cell epitopes have been diversifying throughout the pandemic, with Omicron sub-lineages resulting in the greatest diversification of epitopes. We find that a subset of individuals enriched in HLAs B07:02 and A03:01, are predicted to lose as many as 9 Spike protein epitopes due to Omicron mutations. We aim on validating these findings by both ELISpots and TCR-sequencing using COVID-19-convalescent PBMCs (RECOVER-2 biobank).
Conclusion/Impact: These findings will enable the identification of human subgroups at risk of T cell evasion while shedding light on the viral-host dynamics.

We have constructed a single-sample deconvolution (ssDcon) framework for deconvolving cell-type expression using the existing deconvolution method (CIBERSORTx) and knock-off models. Our motivation is to derive the gene expression profiles (GEP) of the cell types more accurately from bulk tissue samples compared to the existing methods. As a proof-of-concept, we applied the ssDcon framework to publicly available disease datasets and in-house studies, including a Prsostate Cancer Medically Optimized Genome-Enhanced Therapy (PROMOTE) clinical trial. A 20-cell type deconvolution model was constructed using single-cell gene expression counts from 9 metastatic prostate cancer patients from the Boston Bone Metastases Consortium. A balanced model was generated by selecting 500 (or fewer) centroid neighboring cells using the t-sne mapping and 2,220 genes were identified by the mean-dropout rate for the 6,483 representative cells. Then ssDcon framework was applied to construct knock-off models for the 46 bone and 22 soft tissue biopsies (68) from PROMOTE patients. Eleven of the 20 cell types were differentially abundant between the tissue sources, including osteoclasts and osteoblasts enrichment (p= 3.4×10-6 and 4.8 x 10-6, respectively). Furthermore, ssDcon model’s accuracy will be benchmarked against digital spatial profiling and bulk RNA-Seq datasets from 9 bone PDX models.

Spatial omics technologies have made significant advances in recent years. These techniques have improved our understanding of the spatial localization of cells and cell-cell interactions.
Cell-cell interactions are essential for maintaining tissue homeostasis. Understanding the mechanisms of these interactions is crucial because their dysregulation can cause various diseases. Ligand proteins, one of the signaling molecules responsible for cell-cell interactions, bind to cell surface receptors and trigger intracellular signaling pathways. Although the effective range of ligand signaling may play a crucial role in determining the occurrence of cell-cell interactions, it has not been systematically investigated. In this presentation, we developed a new model for the spatial diffusion of ligands from spatial transcriptome data and estimated the spatial range of ligand action. More specifically, we learned the diffusion distance as a parameter by regressing the expression of downstream target genes against the diffused ligand expression using Visium data from human brain tissue. Our model successfully distinguishes between ligands with long and short effective distances, enabling the analysis of cell-cell interactions while considering the effective distance of the ligand.

Biological results: DNA double-strand breaks (DSBs), the most del form of DNA damage, are often mapped using sequencing. However, sequencing data do not provide information on distribution of DSBs among cells, which is crucial for data interpretation. We infer population distribution of DSBs by using computer simulations to integrate replication profiling (progress of replication in cells), quantitative DSB sequencing (average DSB numbers) and gel electrophoresis data (lengths of chromosomal DNA fragments). By simulating DSBs in individual cells, we found that cells react heterogeneously to hydroxyurea-induced replication stress: >95% cells repair DSBs and continue replication, while in <5% cells DSBs remain unrepaired, replication stops and DSBs accumulate. We also use computer simulations to evaluate proposed models of Mus81 cleavage and clarify how Mus81 creates DSBs in Mec1-defficient cells. We conclude that computer simulations are powerful tools infer mechanisms of DSB creation.
Algorithms: The core part of our modeling is simulating DNA replication. To accelerate this calculation and improve memory management we use indexing of genomic locations by “time distance” from nearest replication origin, effectively conducting the simulations in the time-variable. This innovation allows us to simulate human DNA replication and test dependence of fork speed on chromatin context, DNTP availability etc.

The degeneration of substantia nigra dopamine neurons in Parkinson’s disease is characterized in part by perturbations of gene expression networks and aberrant miRNA profiles. Treatment and prognosis of Parkinson’s disease would profit considerably from a correct classification and identification of genetic key drivers and functional regulatory network modules involving both genes and miRNAs that are implicated in PD pathogenesis. We present here a comprehensive integrative analysis of gene expression profiles, miRNA signatures, and other publicly available regulatory databases in order to better understand the collaborative functional role between miRNAs and genes in driving PD. A PD-specific transcription factor (TF)-miRNA regulatory network was generated and its topology and functional impacts upon PD was analyzed. We identified six driver genetic elements (2 genes and 4 miRNAs) and various functional network modules that could conceivably trigger PD etiology. We also uncovered the regulatory links among these key drivers and other genetic elements in the PD network. Finally, neuroprotective agents and other small molecule signatures were predicted to reverse the transcriptional changes caused by the identified functional modules.

Dynamical system-based modeling is pervasive in various areas of biology, e.g. ecology, molecular biology, virology, clinical data modeling. These systems are often studied using methods originating from physics. However, the knowledge of accurately parameterized ODE systems is not realistic. As a result, recently, interest has shifted towards machine learning-based methods.
Prominently, DeepLearning-based approaches have proven successful for time series analysis. However, they are notorious for lack of interpretability and robustness due to their black-box nature. To address this problem, we propose a hybrid neural network model that explicitly embeds expert knowledge. Our approach assumes that a mixed set of time series is generated by a finite set of ODE dynamics, with a known functional form. Our method is designed to simultaneously reconstruct partially observed time series, identify the parameters of the systems, and cluster the data. Clusters naturally arise in biological data, like healthy/diseased patients, cell lines committed to various differentiation paths, or epidemiological dynamics of different virus variants.
This approach provides a valuable tool for dynamical system modeling in biology. Experiments show our model is more interpretable and better at reconstruction than its black-box counterparts. We demonstrate our method using systems describing predator-prey dynamics and oscillatory reactions.

Diagnosing oral diseases like oral submucous fibrosis (OSMF) and oral squamous cell carcinoma (OSCC) is a complex process that requires a trained eye to identify subtle changes in the histological images (HIs). These changes are difficult to detect and often go unnoticed, making accurate diagnosis without the help of a histopathologist almost impossible. However, the shortage of histopathologists and busy schedules cause significant delays in getting a diagnosis. The overlapping features in the HIs between the different disease conditions make it challenging for even experienced histopathologists to make a confident and accurate diagnosis. To overcome these limitations, we developed a deep learning framework that could classify normal, premalignant, and malignant stages; and also differentiate between the three stages of malignant OSCC tissue (well, moderate, and poorly differentiated). The framework was trained and tested on internal datasets consisting of samples from diverse locations in the Indian subcontinent. The results were remarkable, with a validation accuracy of 97%, indicating the effectiveness of the ensemble-based learning approach in grading OSCC. In conclusion, we have shown the potential of using deep learning techniques in a clinical setting to diagnose oral diseases accurately and quickly, overcoming the limitations of traditional methods and improving patient outcomes.

Genome-scale metabolic models (GEMs) provide mechanistic insights into microbial physiology, and are instrumental for numerous biotechnological applications such as rational strain design, and engineering of microbial communities. Among the tools aiming to speed-up GEMs reconstruction, CarveMe stands out for a top-down approach that uses a manually curated universal model of prokaryotic metabolism carved to tailor a particular organism. The last universal model of CarveMe was reconstructed from the BiGG database, being inherently skewed towards model organisms. A new universal model is being developed, transitioning from BiGG to core data resources from ELIXIR (namely Rhea, CheBi, and UniProtKB), with the aim of expanding the previous universal model to capture a broader microbial diversity. This will allow for a more sustainable synchronization of CarveMe with existing resources and improve its ability to generate high-quality reconstructions for non-model organisms.

The immune system acts by mounting a defence that ensures host survival from microbial threats. To engage this faceted immune response and provide protection against infectious diseases, vaccinations are the critical tool developed. However, vaccine responses are governed by levels that separately only explain a fraction of the immune reaction. To address this knowledge gap, we conducted a feasibility study to determine if multi-view modelling can aid in gaining actionable insights and capture the immune system diversity. We thus sought to assess this capacity on the responsiveness to Hepatitis B virus (HBV) vaccination.
Seroconversion to vaccine induced antibodies against HBV surface antigen (anti-HBs) in early-converters and late-converters, was defined based on anti-HBs titres. The multi-view data encompassed bulk RNA-seq, CD4+ T cell parameters, flow cytometry, and clinical metadata while modelling included testing single-view and multi-view joint dimensionality reductions. Multi-view outperformed single-view methods for all metrics, confirming an increase in predictive power. This approach complements clinical seroconversion and all single modalities. Importantly, this modelling could identify what features predict HBV vaccine response, such as age, inflammation-related gene sets and pre-existing vaccine specific T-cells. This methodology could be extended to other vaccination trials to identify key features regulating responsiveness.

We developed a novel tool for reconstructing draft constraint-based metabolic models by “pruning” a universal model, following CarveMe’s general strategy, while addressing limitations in existing tools. Our tool better considers compound transport via passive diffusion, can represent a wider range of pathways than CarveMe, and returns models with minimal blocked reactions – unlike ModelSEED.
The tool accepts the organism’s genome and species name as input, using the latter for taxonomic classification and identification of the core metabolism. The genome is used to predict expressed pathways in the organism. Reactions in the universal model receive scores derived from the alignment score: reactions associated with aligned gene sequences in UniRef90 receive positive scores, while negative scores are assigned otherwise.
The universal model is “pruned” by maximizing the sum of reaction scores while enforcing network connectivity and biomass flux. It generates draft metabolic models sharing more reactions and pathways with manually curated models than those produced by CarveMe or ModelSEED. This demonstrates its potential to improve metabolic model reconstruction, particularly for microbiome simulations, as metabolites can be transported via passive diffusion outside the cell, even if this is not relevant to their own metabolism.

Response prediction of reaction perturbation is an important task in the chemical bioproduction process. One of the methods for response prediction is structural sensitivity analysis (SSA), which needs only metabolic network information. While the benefits of SSA are the theoretical bases and ease of method application, SSA has a limitation that SSA may provide “indefinite” prediction. In this study, we propose a concept of confidence values and BayesianSSA. The confidence values enables interpreting the indefinite prediction of SSA, and BayesianSSA enables reflecting perturbation experiment results in SSA. BayesianSSA can apply a Bayesian update framework, and iterative design-build-test-learn (DBTL) cycles can be easily adopted. We compared performances between BayesianSSA and a base model, which uses only confidence values, on synthetic datasets and the central metabolic pathway of Escherichia coli. As a result, BayesianSSA outperformed a base model, and the results show that BayesianSSA has a high accuracy. We can also see the transition of the response prediction by BayesianSSA. In conclusion, BayesianSSA is a powerful method for the chemical bioproduction, and the Bayesian update framework incorporating BayesianSSA has a potential to accelerate the DBTL cycle iterations.

Disease recurrence and therapy resistance are still leading challenges for patients despite advancements in cancer treatments. Genetic and epigenetic changes provide tumour cells the ability to continuously proliferate and to escape from apoptosis and growth suppression, causing resistance to monotherapy-based interventions. Combination therapies have promise to offer higher efficacy by simultaneously targeting compensatory mechanisms altered by tumours, and reducing toxicity through the use of lower doses. In this project, we aim to identify topological relationships between drug targets and the disease contexts in cell line specific protein protein interaction networks. To accomplish that, we first combined several independent combinatorial drug screen datasets to generate one of the largest datasets of its type. We, then, integrated cell line specific multi-omics data to model differential drug responses to elucidate sensitivity and synergy related biomarkers, and to reconstruct cell line specific subnetworks. Ultimately, this project promises to identify novel molecular biomarkers, and elucidate context specific relationships between drug targets and found biomarkers in the case of synergy and sensitivity.

The growing volume and complexity of omics data has opened the door to a system-wide understanding of biological processes. Popular approaches to exploit the ensuing datasets include differential expression and co-expression network analyses. While both yield insights about biological entities and pathways involved in those processes, they fail to capture underlying mechanisms.
We developed a workflow to extract information from both existing knowledge and experimental data, and build a dynamic network of binary variables to elucidate the modus operandi of MVA, a vaccine approved against monkeypox and smallpox. The resulting model was calibrated to mimic the gene expressions and cell abundances in three immunized animals.
Despite the approximation inherent to the Boolean formalism and the stringent statistical filtering, the network built from four immune pathways and eleven cell populations successfully recapitulated the observed dynamics of those populations following immunization (reported in Rosenbaum et al., Front. Immunol. 2018). More importantly, it gave insights about the interacting genes responsible for these dynamics, including the underlined shared behavior of the granulocytes and monocytes subsets on one hand, and lymphocytes on the other.
This model will allow us to make mechanistic hypotheses for MVA-induced inflammatory responses to be validated experimentally in a 3R-motivated approach.

Glioblastoma is the most aggressive form of brain tumor, currently lacking an effective cure. Its resistance mechanisms are rooted in tumor heterogeneity and plasticity, which permit reversible transitions between tumor cell phenotypes. Advanced mathematical models are crucial for understanding tumor biology and formulating effective treatment strategies. In this study, we constructed a mathematical model of glioblastoma cell population dynamics using Itô’s stochastic differential equations. We parametrized the model using our in-house single-cell RNAseq data and cell growth data. The initial cell distribution across various cell cycle states was estimated using the deconvolution method implemented in the consICA package. By examining cell growth dynamics, we formulated a quality criterion and performed parameter fitting using the particle swarm method. This allowed us to compute a vector of the model’s parameters and successfully align the model with empirical data. We also estimated the evolution of cell proportions at each cell cycle stage throughout the observation period. Our model quite accurately replicates the biological process of cell proliferation, taking into account random factors influencing it. Adaptable to experimental data, the model maintains its interpretability, offering a robust mathematical representation of cancer cell population dynamics.

Bacterial metabolism can be mathematically represented as a large network of chemical reactions. In such networks, propagating perturbations can carry information about environmental perturbations. To quantify this information, we need to study the responses of bacterial components in the presence of noise. Metabolic Control Analysis is a powerful framework for analyzing metabolic responses and characterizing the relationship between steady-state properties. Here we extend this framework to study the system response to perturbations following probability distributions. Mutual information between model variables is used to quantify their dependencies as a form of information transfer. In an exploratory study, we considered small metabolic networks with different types of enzyme regulation and studied information transfer via static or periodic perturbations. We plan to extend this approach to larger models and to computationally optimize the information transfer through the metabolism of bacteria, using mutual information between specific cell variables as an optimization objective. We expect this work will help better understand the signal-processing capacities of bacteria, taking into account internal and environmental noise and uncertainties.

Computational methods combining the mathematical formulation and simulation of multi-layered connections with interactive visualization solutions are in the first stages of development. The main feature making this concept recognizable from the biological community is the option to inspect the system as a whole, even given the different granularity levels of its components. We propose a new modeling paradigm that allows modelers to integrate fine- and coarse-grained biological information into unified models. Due to its clinical importance, we exploited the paradigm to model Clostridium difficile metabolism during infection. Mechanistic models provided by the -omics adaptation of Flux Balance Analysis (FBA) have extended this concept to the analysis of metabolic regulation. However, the predictions offered by FBA can be strongly affected by flux boundaries (in particular, fluxes of reactions that sink nutrients). To deal with uncertainty, we introduced an open-source and general modeling framework based on a graphical meta-formalism to simplify the modeling phase. The proposed paradigm implementing two solution techniques (i.e. ODEs and FBA) can capture the levels of system granularity. Our framework allows performing functional studies where the understanding of the multi-level stable condition of the system in fluctuating conditions is combined to investigate the functional dependencies among entities.

The SysMod COSI organizes annual gatherings at ISMB. This short talk will introduce all speakers, organizers, and the main topics of the 2023 meeting. This year’s meeting incorporates four sessions covering, beginning with “Modeling Biological Systems from Micro- to Macro-Scale,” followed by “Integrative Modeling Approaches for Biological Systems” and “Computational Modeling of Diseases.” It will conclude with a session on “Modeling Metabolic and Dynamic Tumor Evolution.” Two outstanding keynote speakers will present their visions on developments in these fields: Ina Koch from the Goethe University Frankfurt am Main, Germany, and Thomas Höfer from the German Cancer Research Center (DKFZ), Heidelberg, Germany. Chiara Damiani will close the event by awarding this year’s poster prizes. The event is hosted by Matteo Barberis and Chiara Damiani.

The increasing amount of available experimental data enables us to consider biological systems at multi-scale levels. Different computational methods, ranging from discrete techniques to differential equation-based methods, have been developed to build models at different scales depending on the experiment. The talk addresses the possibilities and challenges of the application of Petri nets to analyze biological systems. To illustrate the different applications, we present in-silico knockout studies of xenophagic capturing of Salmonella and analysis of cellular processes in the lymph node. Finally, the combination of Petri nets with agent-based modeling will be discussed as a future direction.

The organization of biological systems as distinct but connected layers poses a grand challenge for biomathematical modeling, because processes occurring at the various layers have different time scales and almost always focus on different types of variables. The higher layers usually correspond to a “big picture” of physiological events, whereas the lower levels account for increasing granularity and detail. When investigating a system at a high level, it is infeasible to carry along all details from lower levels, partly for technical reasons, but more so because they would overwhelm insights at the higher level due to their sheer numbers and the fact that they typically run on much faster time scales. We address this situation with the first application of “Template-and-Anchor modeling” in the proposed implementation. A template is a high-level model that focuses exclusively on the main physiological components. Anchor models provide specific biological details characterizing the mechanisms that govern the system and are represented in the template model as variables. We use this approach to investigate the effect of dioxin exposure on human health. By adjusting parameter values within the anchor models, the overall template model can be personalized, thereby offering the option of personal health risk assessments.

This study addresses the need to understand better human embryonic development to improve assisted reproductive technologies such as in vitro fertilization. Novel technologies such as transcriptomics can provide single-cell level data to understand embryo development from a genetic and metabolic point of view. The study aims to develop a computational model to discriminate different developmental stages during trophectoderm (TE) maturation using scRNAseq data. The proposed method involves selecting pseudo-perturbations specific to each developmental stage, allowing for learning Boolean network models. These models are inferred from the pseudo-perturbations and prior-regulatory networks and optimally fit scRNAseq data for each developmental stage. The main result is the proposal of a general framework for inferring Boolean networks from scRNAseq data. Another result is identifying a family of Boolean networks specific to medium and late TE developmental stages, revealing opposite regulation pathways and supporting biological hypotheses in this domain.

The cell cycle is a highly regulated process that ensures the accurate replication and transmission of genetic information from one generation of cells to the next. We devised a quantitative description of gene expression dynamics during the cell cycle in mouse embryonic stem cells. We combined high-depth single-cell multiomics sequencing, biophysical modeling, and advanced deep learning techniques to develop a novel method that allows us to infer cell cycle dependent gene expression dynamics. We performed multiome sequencing, namely scRNA-seq and scATAC-seq in mouse embryonic stem cells. We used a generative deep learning tool that assigns a latent cell-cycle phase to the cells based on the spliced and unspliced mRNA reads. Using this latent cell-cycle phase, we developed a biophysical model that describes the dynamics of gene-specific mRNA synthesis and degradation rates during the cell cycle. Our model helped to identify key regulators that drive the transcriptional dynamics during the cell cycle. By extending this approach to scATAC-seq, we were able to investigate the chromatin accessibility during cell cycle progression.

The animal testing ban for safety evaluation of cosmetic ingredients urges the need for new approach methodologies’ development, mainly to better understand xenobiotic metabolic Mechanisms of Action (mMoA) for systemic toxicity assessment. To this end, we developed a workflow to combine endogenous metabolic knowledge from a Genome Scale Metabolic Network (GSMN) and in vitro transcriptomics data by building condition-specific metabolic networks, which involves 3 main steps. The first step consists in building condition-specific models representing the metabolic impact of each exposure condition, while taking into account the diversity of possible optimal solutions with a partial enumeration step. Then, 2 conditions, represented by 2 sets of several optimal condition-specific networks, are compared by extracting differentially activated reactions (DARs) between these 2 sets. Finally, using distance-based clustering and a subnetwork extraction method, DARs are grouped into clusters of functionally interconnected metabolic reactions. The workflow was applied to two well-known hepatotoxic molecules, amiodarone and valproic acid. Despite large disparities in evidenced transcriptomic effects for these two chemicals i.e., 2 DEGs for Amiodarone and 5709 DEGs for Valproic Acid, we were able to model and visualize different mMoA, fitting several evidence in the literature.

Single-cell time series data frequently display considerable variability across a cell population. The current gold standard for inferring parameter distributions across cell populations is the Global Two Stage (GTS) approach for nonlinear mixed-effects models. Although the GTS method is reliable, its current implementation requires repeated use of non-convex optimization, which is not guaranteed to converge, while each optimization run requires multiple simulations of the system. We propose an alternative, computationally efficient implementation of the GTS method for mixed-effects dynamical systems which are nonlinear in the states but linear in the parameters (a class that encompasses a wide range of models such as those based on mass-action kinetics). For such systems, point parameter estimates can be obtained using least squares regression on time derivatives of smoothed measurement data, an approach called gradient matching. Here, we extend the application of gradient matching to the inference problem for mixed-effects dynamical systems and integrate it into the GTS method by properly accounting for uncertainties in individual cell parameters in the first stage. We also present an Expectation Maximization (EM) algorithm and associated parameter uncertainty estimates which are applicable when not all system states are observed, as is typically the case for biological systems.

Quantitative dynamical models facilitate the understanding of biological processes and the prediction of their dynamics. However, these models typically have unknown parameters that need to be estimated from experimental data. While there are various methods and software tools available for quantitative data, the options for semi-quantitative and qualitative data are limited and computationally demanding.
To address this challenge, we propose a novel approach that integrates censored, ordinal, and nonlinear-monotone data into the parameter estimation process using a combination of optimal scaling and spline modeling approaches. To integrate ordinal and censored data, we use the optimal scaling approach, which involves representing qualitative data as quantitative surrogate data that accounts for constraints on their relation. For nonlinear-monotone data, we optimize splines to model the unknown data dependencies. These approaches enable us to treat the data as if it were quantitative, such that we can use pre-existing software parameter estimation pipelines.
To improve the method’s efficiency, we formulate the inverse problem as a bi-level optimization problem and compute gradients using an efficient semi-analytical algorithm. We apply it to a model with all four data types and compare the results. The approach is implemented in the open-source Python Parameter Estimation TOolbox (pyPESTO)

Several studies attempted to link genotype-phenotype relationships yet it remains unclear how genetic interactions between quantitative trait nucleotides (QTNs) can drive phenotypic variation. If QTNs modulate phenotypic variation in a metabolically driven process, it is obvious to ask: how do these QTNs individually and in combinations change the connectivity of metabolic regulators? Furthermore, how does metabolic flux distribution change as QTN interacts? To test our hypothesis, we harness the gene expression data obtained from an allele replacement panel of Saccharomyces cerevisiae and study how QTNs in the three genes: two coding polymorphisms in IME1 and RSF1 and two non-coding polymorphisms in RME1 and IME1, can modulate sporulation efficiency variation. Using differential gene expression analysis and network analysis we show several metabolic regulators change connectivity as QTNs interacts. We integrated the gene expression data of each QTN combination into genome-scale metabolic models to reconstruct QTN-specific metabolic models. Using genome-scale differential flux analysis we observed flux variation in the amino acid biosynthesis pathway, pentose phosphate pathway, and glycerophospholipid metabolism as a consequence of QTN-QTN interactions. The underlying principles gained from this study can be anticipated for complex human diseases where multiple SNPs can interact and contribute to a disease phenotype.

Infections caused by multiple pathogens, known as poly-microbial infections, can worsen patient prognosis. Acinetobacter baumannii and Staphylococcus aureus are two bacterial pathogens frequently co-isolated in infections. Both belong to the ESKAPE group, which is associated with high rates of antimicrobial resistance, and are responsible for the majority of nosocomial infections. However, the interaction between these two pathogens during co-infection remains poorly studied.
Therefore, we implemented an extended logistic growth model based on ordinary differential equations to quantitatively compare the growth parameters of the two species in different experimental settings. Experiments were performed using a variety of different laboratory strains as well as clinical isolates for both species in order to identify the key mechanisms of their interaction, while taking into account the biological variation observed in the clinics. In addition, wild-type strains and specific knock-out mutants were co-cultured and grown separately in the supernatant of the other strain to elucidate contact-dependent and contact-independent processes. Calibration of the model using this big volume dataset revealed a complex network of interactions between the species, involving both cooperative and competitive elements. This systems biology approach advances our understanding of co-infection processes and paves the way for developing improved treatment strategies.

Our systems biology approach integrates mathematical modeling, multi-omics data, and molecular signaling networks to gain a comprehensive understanding of biological systems. This framework was applied to Alzheimer’s disease (AD) to develop a model of insulin signaling in the blood-brain barrier (BBB) and its impairment in metabolic syndrome and AD. The model was based on ordinary differential equations (ODEs) and encompassed two interrelated subsystems: insulin signaling transduction in BBB endothelial cells and turnover of vascular-cell adhesion molecule 1 (VCAM1), a marker for cerebrovascular inflammation. The model was validated using western blot and proteomics data and applied to an AD patient and control RNA-Seq data. The in-vitro findings showed that insulin stimulation triggered the phosphorylation of various targets in a time- and dose-dependent manner and VCAM1 expression was reduced by insulin treatment. The mechanistic model successfully described the experimental results and predicted potential signaling perturbations due to amyloid beta (Aβ) exposure. The model was also used to examine transcriptomic changes in individuals with AD, identifying molecular mediators contributing to BBB dysfunction in AD. This study shows a systems biology model accurately representing the insulin signaling cascade and downstream expression of VCAM1 in BBB. The model can identify potential therapeutic targets for AD.

BRAF is prototypical of oncogenes that can be targeted therapeutically and the treatment of BRAFV600E melanomas with RAF and MEK inhibitors results in rapid tumor regression. However, drug-induced rewiring generates a drug adapted state thought to be involved in acquired resistance and disease recurrence. In this study, we investigate mechanisms of adaptive rewiring in BRAFV600E melanoma cells using an energy-based implementation of ordinary differential equation (ODE) modeling in combination with proteomic, transcriptomic and imaging data. We develop a method for causal tracing of ODE models and identify two parallel MAPK reaction channels that are differentially sensitive to RAF and MEK inhibitors due to differences in protein oligomerization and drug binding. We describe how these channels, and timescale separation between immediate-early signaling and transcriptional feedback, create a state in which the RAS-regulated MAPK channel can be activated by growth factors under conditions in which the BRAFV600E-driven channel is fully inhibited.

We introduce SMITH – Stochastic Model of Intra-Tumor Heterogeneity – a novel approach to computational modelling of cancer cell populations and their evolution. SMITH introduces the concept of “confinement”, a mathematical representation of growth constraints within a foundational branching model of cancer development. Using this confinement mechanism, SMITH can emulate the heterogeneity observed in various cancer types with distinct spatial structures, such as breast cancer or lymphoma. In doing so, we achieve comparable outcomes to results produced by more intricate cellular-automata-based models. However, in contrast to cellular automata, the simplicity of our form of branching process model permits the simulation of realistically-sized tumours of up to one billion cells. To showcase the efficacy of SMITH, we performed over 10,000 simulations with a billion cells each. We then used a point cloud distance minimization over our simulation results to obtain parameters matching the different tumour types in both their mutation load and clonal diversity. Our analyses show that the confinement mechanism is sufficient to reproduce commonly observed evolutionary patterns and clonal dynamics.

In recent times, to investigate the interplay dynamics between immune and tumor cells in human cancer, several computational modeling methods like agent-based models have been employed. However, since each tumor has its unique tumor microenvironment (TME), a personalized and specialized study of each cancer case is necessary.
In this perspective, we introduce MAST, which is a hybrid Multi-Agent Spatio-Temporal model that reproduces specific TME scenarios starting from high-throughput sequencing data. The integration of an agent-based model with a continuous partial differential equations (PDE) model, enables the inclusion of crucial aspects of the tumor microenvironment. This encompasses the spatio-temporal nature of cancer progression, its reliance on the availability of nutrients, the immune response, as well as the development of mutation-based mechanisms that lead to evasion. The proposed approach was tested by simulating four human colorectal cancer subtypes starting from genomics and transcriptomics data, coming from both bulk and single-cell sequencing technologies, of human colorectal cancer tissue. Both the emergent properties of the four simulated TMEs and the spatial and temporal evolution of the four TME specific in-silico cancer progression largely agree with the current biological knowledge and patient outcomes, thus supporting the validity of the approach.

Epithelial ovarian cancer involves forming spheroids responsible for disease metastasis, recurrence, and lower chances of recovery. Although cancer progression has already been linked with metabolic differences in tumor cells, possible associations between metabolic landscape and metastatic morphological transitions remain unexplored. The present study aimed to identify metabolic perturbations during the phenotypic shifts of three distinct morphologies (2D monolayers and two geometrically individual three-dimensional spheroidal states) of the high-grade serous ovarian cancer line OVCAR-3. We performed quantitative proteomics and integrated protein states onto genome-scale metabolic models to construct context-specific metabolic models for each morphological stage of the OVCAR-3 cell line. Using these models, we obtained metabolic reaction modules responsible for disease progression and determined gene knockout strategies to reduce metabolic alterations associated with disease progression. The DrugBank database was explored to mine drugs and evaluated their effect in impairing metastatic morphological transitions. Finally, we experimentally validated our predictions by confirming the ability of one of our predicted drugs: the neuraminidase inhibitor oseltamivir, to disrupt the metastatic spheroidal morphologies without any cytotoxic effect on untransformed stromal mesothelial monolayers. The current work expands our horizon on ovarian cancer progression and provides a methodological framework to identify novel targets against cancer progression.

Somatic evolution is a complex process shaped by the interplay of stem and progenitor cell dynamics, mutation and selection. None of the associated parameters can be directly measured in humans. In the first part of my talk, I will discuss inference approaches to reconstruct the evolution of tumors from genomic sequencing data. Focusing on a tumor of early childhood, neuroblastoma, I will show how insight into tumor evolution might help improve treatment decisions. A key insight of this work is the interplay between stem/progenitor cell dynamics on the hand, and the occurrence and fixation of driver mutations on the other hand. In the second part, I will discuss how mathematical analysis of this interplay has supported engineering tumorigenesis in mice without introducing oncogenes.

This remark will provide a brief overview of the SysMod meeting, covering the speakers, chairpersons, and organizers. The event consisted of four sessions that delved into diverse methodologies and applications of computational systems modeling in the field of biology. We were privileged to have two distinguished keynote speakers, Ina Koch and Thomas Höfer, who delivered insightful talks. We express our gratitude for all the valuable contributions made to the scientific posters and we recognize and honor the best ones through the annual SysMod poster awards in 2023.