Computational modelling in health and disease: highlights of the 6th annual SysMod meeting

Abstract

Summary

The Community of Special Interest (COSI) in Computational Modelling of Biological Systems (SysMod) brings together interdisciplinary scientists interested in combining data-driven computational modelling, multi-scale mechanistic frameworks, large-scale -omics data and bioinformatics. SysMod’s main activity is an annual meeting at the Intelligent Systems for Molecular Biology (ISMB) conference, a meeting for computer scientists, biologists, mathematicians, engineers and computational and systems biologists. The 2021 SysMod meeting was conducted virtually due to the ongoing COVID-19 pandemic (coronavirus disease 2019). During the 2-day meeting, the development of computational tools, approaches and predictive models was discussed, along with their application to biological systems, emphasizing disease mechanisms. This report summarizes the meeting.

Availability and implementation

All resources and further information are freely accessible at https://sysmod.info.

1 Introduction

SysMod is the community for data-driven computational modelling and multi-scale analysis of biological systems within the International Society for Computational Biology (ISCB) (Dräger et al., 2021). The 6th annual community meeting took place in a virtual setting as part of the yearly Intelligent Systems for Molecular Biology (ISMB) conference in 2021.

2 Day 1

2.1 Session I: Disease and multi-scale modelling—part I

The first session, moderated by Dr Claudine Chaouiya, opened with the keynote talk ‘Whole-body metabolic modelling provides novel insight into host-microbiome crosstalk’, delivered by Dr Ines Thiele. Models cover metabolism, physiology, diet together with the gut microbiome and thus constitute valued tools to assess metabolic diseases. Dr Thiele emphasized manually curated and annotated knowledge bases and high-quality genome sequences to develop mechanistic models (Heinken et al., 2021; Heinken and Thiele, 2022). Dr Thiele and colleagues develop and maintain the virtual metabolic human database that assembles information about human and gut microbial metabolism and links this information to disease and nutritional data (Noronha et al., 2018).

Next, Dr Reihaneh Mostolizadeh presented workflows for modelling microbial community interactions applied to Dolosigranulum pigrum and Staphylococcus aureus within the human nose. She highlighted the need for better implementations of host–pathogen interactions models to mimic the human environment and showcased how the presented method can identify new ways to inhibit disease-causing infections through microbe–microbe interactions (Glöckler et al., 2022). The first session ended with a talk delivered by Paul F. Lang on BpForms and BcForms, toolkits for concretely describing non-canonical polymers and complexes (Lang et al., 2020). The speaker then presented his efforts to use the toolkits to build global models of whole cells.

2.2 Session II: Disease and multi-scale modelling—part II

The second session, moderated by Dr Anna Niarakis, opened with Marco Ruscone presenting a model developed in PhysiBoSS (Letort et al., 2018), a tool that combines two modelling approaches, PhysiCell (Ghaffarizadeh et al., 2018) to perform agent-based simulations of cell populations, and MaBoSS (Stoll et al., 2017) to perform stochastic simulations of Boolean models of signalling pathways. PhysiBoSS allows the study of the different cell migration modes (single cell, collective or trail) and reproduces experimental observations by simulating how intracellular alterations impact tumour growth. The following talk, delivered by Eirini Tsirvouli, focused on the logical and experimental modelling of keratinocytes in psoriasis to investigate treatment options. The speaker emphasized the immunomodulatory role of CD4+ T-helper 1 and T-helper 17 cell cytokines and how they promote the development of psoriasis. The response to treatment with a cPLA2 inhibitor or Calcipotriol revealed a distinct mode of action for both drugs demonstrating how the study of complex diseases can benefit from integrated systems approaches (Tsirvouli et al., 2021). The next talk by Cedric Lhoussaine presented the limitations and perspectives of a glucose-insulin model that investigates intestinal absorption in type 2 diabetes. This work built on a well-known model of postprandial glucose response (Dalla Man et al., 2007), based on partial parameter estimation, and investigated the capability of this model to predict which physiological compartments were most likely able to restore a normal glycaemia from a pathological one in two different datasets (Dursoniah et al., 2021). The last talk of the session was delivered by Dr Lal Puniya, who discussed the development of constraint-based models of genome-scale metabolism in CD4+ T cells and their use to identify novel targets for drug repurposing in autoimmune diseases. The metabolic behaviours were combined with gene expression data of three autoimmune diseases, rheumatoid arthritis, multiple sclerosis and primary biliary cholangitis, resulting in the identification of 68 metabolic drug targets (Puniya et al., 2021).

2.3 Round table discussion

The morning sessions ended with a lively round table discussion, moderated by Dr Niarakis, including Dr Boris Kholodenko, Dr Claudine Chaouiya and Dr Marc Birtwistle as panel guests. The speakers first discussed the difficulties in bridging different scales spanning distinct biological information for which a wealth of -omics data is being generated. Dr Kholodenko emphasized the usefulness of perturbation-based methods to study local interactions in the signalling cascades. A second discussion focused on the uncertainty arising from incomplete datasets and fragmented knowledge about the biological mechanisms associated with health and disease. Dr Birtwistle elaborated on how we can fill in the missing causal pieces by making networks context-specific and paying attention to feedback loops that might exist in different cellular contexts and impact the system’s behaviour. All speakers agreed on the need to use standards to enhance interoperability and reproducibility and highlighted the need to address data sparsity. Dr Birtwistle noted that while we are in the era of big data, access to the suitable type of data needed for identifying causal and cell-context-specific links is not straightforward. This would include perturbation and time-course data. Dr Chaouiya added the importance of recognizing what is missing from our models and acknowledging their limitations. The third question focused on the challenge of modelling the virtual human. Dr Kholodenko commented on the importance of setting ambitious goals, such as building digital twins, to drive the field forward, even if all challenges have not been addressed yet. As a growing field of research, goals and perspectives need to be complemented with a dedicated experimental design to generate valuable data.

2.4 Session III: Infectious disease modelling

The third session of the day started with a talk by Jordan Weaver, who presented a multi-cellular spatial model for investigating the factors controlling plaque growth dynamics. The model predicted that antiviral responses in lung epithelial cells exhibit either linear radial growth or arrested plaque growth depending on the local concentration of type I interferons (Aponte-Serrano et al., 2021). The next talk presented by Lauren Benoodt discussed STREGA-NONA, a genetic algorithm implementation for identifying gene-set associations in networks using single-cell transcriptomics (Katanic et al., 2016; Thakar et al., 2020). STREGA-NONA integrates knowledge with data to identify gene-set specific topology and differential activity with single-cell RNAseq data. Using mutual information, analysis of subnetworks of naive T-cell clusters identified associations among genes related to an inflammatory response in HIV infection. The session concluded with two talks on the harmful pathogen Pseudomonas aeruginosa PA14, which often possesses multiple resistance mechanisms against antibiotic agents. First, Dawson Payne introduced iPau21, the latest iteration of the genome-scale metabolic model of P. aeruginosa, which characterizes mucin-driven shifts in bacterial metabolism, highlighting the importance of contextualized models to recapitulate known phenotypes of unaltered growth and differential utilization of fumarate metabolism Payne et al. (2021). Next, Sanjeev Dahal presented a model to provide mechanistic explanations for two case studies: (i) effect of gene deletion on gluconate production; and (ii) metabolic influence on drug tolerance of P.aeruginosa (Dahal et al., 2021). Both talks proposed computational frameworks to aid the development of effective intervention strategies against P. aeruginosa.

2.5 Poster session

The SysMod meeting had 30 posters describing various approaches for mechanistic modelling, including agent-based, constrained-based and logical models. Presented approaches also included stochastic and deterministic methods to integrate data and improve the knowledge base contents.

3 Day 2

3.1 Session IV: Integrative approaches and methodologies—part I

The second day started with a session moderated by Dr Laurence Calzone. The opening talk was delivered by the second keynote presenter of the meeting, Dr Ruth Baker, titled ‘Identifiability and inference for models in mathematical biology’. Dr Baker emphasized the use of simple models to understand complex biology. The talk described Bayesian and ordinary differential equation (ODE) models to study the regulation of RNA-protein complexes and mRNA expression in Drosophila. In addition, a stochastic cell motility model using electrotaxis data was developed to predict the role of preferential polarizing in cell motility using a Bayesian approach (Prescott et al., 2021). The take-home message of this talk was how to develop models that are identifiable and can be efficiently calibrated to quantitative data. The following talk delivered by Nantia Leonidou focused on identifying potential severe acute respiratory syndrome coronavirus type 2 (SARS-CoV-2) inhibitors through metabolic perturbations in infected cells. She presented a re-implementation of the mCADRE algorithm in Python that generates tissue-specific metabolic models Leonidou et al. (2022). The last talk of the session was delivered by Melania Nowicka, who discussed the design of distributed cell classifier circuits with genetic algorithms and logic programming. The speaker showed evaluation results on actual cancer data demonstrating that distributed classifiers outperform single-circuit designs.

3.2 Session V: Integrative approaches and methodologies—part II

The fifth session of the meeting, moderated by Dr Matteo Barberis, opened with the talk of Miroslav Phan, who presented REGIR, a rejection-based Gillespie algorithm to efficiently simulate non-Markovian stochastic processes. The algorithm was applied to model a public mouse stem cell dataset and faithfully recapitulated the underlying biological processes. Emilee Holtzapple delivered the subsequent talk and presented a data-driven Glioblastoma stem cell model used to study cell line differences regarding treatment resistance. The model was fitted to gene expression data to better predict individual cell line responses to treatment and could successfully predict a high percentage of kinase inhibitor treatments in a cell-line specific manner. On a similar integrative vein, next Maurício Moreira-Soares presented a talk focusing on a phase-field model where cells are described as droplets with surface tension that interact with the extracellular matrix by adhesion and excluded volume. An equivalent system using a Dissipative Particle Dynamics (DPD) approach was produced, observing that adhesion regulates cell deformability and enhances migration under extreme confinement conditions for both models. In the next talk, Marzia di Filippo described a model-based data integration pipeline to characterize the multi-level regulation of cell metabolism. Differential reaction expression was computed from transcriptomic data, and constraint-based modelling was used to predict whether differential expression of metabolic enzymes directly resulted in differences in metabolic fluxes. In parallel, predictions were made about how differences in substrate availability translate into differences in metabolic fluxes using metabolomic data. The fifth session ended with a talk delivered by Dr Matteo Barberis, who emphasized the use of a simple model to rationalize complex dynamic responses. Dr Barberis focused on the evolutionarily conserved wave-like pattern of mitotic cyclin-dependent kinase (Clb/Cdk1) enzymatic activities, which guarantee coordination of DNA synthesis with cell division, thus cell proliferation. This oscillatory behaviour was investigated in a minimal model of the Clb/Cdk1 network for the eukaryotic model organism budding yeast. By integrating computational prediction and experimental testing of the minimal network, a novel regulatory design was unravelled that involves a Clb-transcription factor axis, which robustly sustains Clb/Cdk1 oscillations (Barberis, 2021a,b). This novel design principle rationalizes the quantitative model of Cdk control proposed by the 2001 Nobel Prize recipient Sir Paul Nurse.

3.3 Session VI: Structure-based dynamic modelling and SysMod poster award 2021

The last session of the workshop hosted the third keynote talk, entitled ‘Structure-based dynamic modelling reveals ways to overcome kinase inhibitor resistance and oncogenic RAS signalling’, delivered by Dr Boris Kholodenko. Dr Kholodenko described a rule- and structure-based model of the Ras to ERK cascade and its application to drug treatment. By simulating adaptions to the network, these models enable investigators to investigate potential drug treatments. The modular analysis, in particular, allows global analysis in terms of local analysis and can be used for network reconstruction. The meeting concluded with remarks from Dr Juilee Thakar, who summarized the pillars of mechanistic modelling, namely high-quality experimental knowledge, rigorous parameter, perturbation analysis and standardization of computer models for reproducibility. Current roadblocks include (i) the lack of algorithms to improve the annotations of high-throughput datasets and the development of mechanistic models from data-intensive studies. Machine-learning tools are expected to help address some of these roadblocks; (ii) challenges applying mechanistic approaches to human data in the absence of time-course and perturbation data; and (iii) the mechanistic modelling is limited by highly time-consuming literature curation.

4 Outlook

Future SysMod meetings will likely be organized in a hybrid model that facilitates remote and in-person participation. The 7th SysMod meeting was the first meeting in a hybrid setting during the ISMB 2022 conference in Madison, Wisconsin, USA.

Data Availability

For further general and up-to-date information about SysMod, we refer the reader to https://sysmod.info.