TISON: a next-generation multi-scale modeling theatre for in silico systems oncology

Multi-scale models integrating biomolecular data from genetic, transcriptional, and translational levels, coupled with extracellular microenvironments can assist in decoding the complex mechanisms underlying system-level diseases such as cancer. To investigate the emergent properties and clinical translation of such cancer models, we present Theatre for in silico Systems Oncology (TISON, https://tison.lums.edu.pk), a next-generation web-based multi-scale modeling and simulation platform for in silico systems oncology. TISON provides a “zero-code” environment for multi-scale model development by seamlessly coupling scale-specific information from biomolecular networks, microenvironments, cell decision circuits, in silico cell lines, and organoid geometries. To compute the temporal evolution of multi-scale models, a simulation engine and data analysis features are also provided. Furthermore, TISON integrates patient-specific gene expression data to evaluate patient-centric models towards personalized therapeutics. Several literature-based case studies have been developed to exemplify and validate TISON’s modeling and analysis capabilities. TISON provides a cutting-edge multi-scale modeling pipeline for scale-specific as well as integrative systems oncology that can assist in drug target discovery, repositioning, and development of personalized therapeutics.


Introduction
complex inter-, intra-, and extracellular biomolecular regulations that give rise to system-23 level effects (1,2, 18). 24 Rapid advancements in molecular biology, particularly in high-throughput 25 genomics, transcriptomics, proteomics, and metabolomics have generated a vast amount 26 of complex spatiotemporal data in both physiological and pathological contexts (19,20). 27 Such experimental data now populates several online expression databases e.g. 14 However, its use of the cellular potts model increased the computational cost for large- 15 scale models while the calibration of the Monte Carlo time step to physical time step 16 makes its employment in multi-scale modeling challenging (44). In comparison, 17 ELECANS (Electronic Cancer System) (45) provided a feature-rich modeling platform for 18 building multi-scale models to decode the multifactorial underpinnings of tumorigenesis. 19 The platform's Software Development Kit (SDK), however, also placed a heavy C# 20 programming requirement for its users. Similarly, CHASTE (Cancer Heart and Soft Tissue 21 Environment) (46) required test-driven development by using several mathematical 22 modeling frameworks for solving Ordinary and Partial Differential Equations 23 (ODEs/PDEs). Like ELECANS, the programming skills required for using CHASTE also 24 hindered its utilization by conventional wet-lab biologists and clinicians. R/Repast (47), a 25 recently reported platform, provided a High-Performance Computing (HPC) capability 26 towards greater scalability but lacked a programming interface for implementing 27 subcellular biomolecular models. Nevertheless, the lack of a generic and intuitive 28 software providing a "zero-code" modeling environment continues to impede the 29 development and employment of complex multi-scale biological models in research 30 laboratories and clinical settings (48). 31 In this work, we propose a next-generation web-based multi-scale modeling 32 platform "TISON" -Theatre for in silico Systems Oncology. TISON provides a "zero-code" 33 modular environment, which is conveniently employable by modelers, experimental 34 6 biologists, and clinicians, alike. The software comprises of eight scale-specific editors: (i) 1 the Networks Editor (NE) allows for construction and analysis of rules and weight-based 2 biomolecular networks, (ii) Therapeutics Editor (TE) helps develop therapeutic screens 3 on biomolecular networks developed using NE towards identification of novel drug 4 targets, drug repurposing, and personalized therapeutics, (iii) Environments Editor (EE) 5 assists in the creation of diffusive microenvironments towards modeling the dynamical 6 engagement of environmental cues with cellular organoids, (iv) Cell Circuits Editor (CCE) 7 helps construct cell decision circuits as Finite State Machines (FSMs) (49) for computing 8 the cell fate outcomes in light of biomolecular network regulation and microenvironment, 9 (v) Cell Lines Editor (CLE) then assigns these cell circuits to in silico cell line models, (vi) 10 Organoids Editor (OE) employs in silico cell lines to create tissue organoid systems, (vii) 11 Simulations Editor (SE) simulates the tissue organoids to investigate their spatiotemporal

22
Theatre for in silico Systems Oncology (TISON) platform is designed using the distributed 23 three-tier software architecture consisting of components organized into front-end, 24 middleware, and back-end. The front-end consists of eight web-based graphical user 25 interfaces (GUIs) termed "editors". Each editor provides scale-specific modeling features 26 and setting up of associated parameters towards a scale-by-scale development of 27 systems oncology models. The resulting models are taken up by the middleware that 28 consists of a high-performance simulation engine, which has been implemented as a three-layer application comprising of the front-end, middleware, and back-end. The front-5 end includes a web application that contains eight editors with corresponding GUI's. The 6 middleware consists of controllers and models, which take user-defined parameters from 7 the GUIs, compute the simulation logic in light of model parameters, and store the results 8 in the database at the back-end. The back-end stores serialized data in a Microsoft SQL 9 database for provision to the middleware for onward processing.

10
TISON users can begin the model construction process by creating a project followed by 11 defining the world size (see Supplementary Material, Section 1) following which scale-12 specific modeling components can be developed in TISON editors. In the following sub-  Networks Editor -Design and analysis of biomolecular regulatory networks 17 The process of multi-scale modeling in TISON begins with the construction of 18 biomolecular networks using Networks Editor (NE). Users can choose between creating 19 (i) rules-based, or (ii) weight-based network models for onward analyses. For rules-based 20 networks, NE allows defining of Boolean rules as abstractions of biomolecular regulation. 21 Deterministic analysis (DA) (54) can then be performed on these networks towards 22 investigating their regulatory dynamics and cell fate outcomes. Results obtained from DA 23 can be visualized as cell fate and attractor landscapes (51,55). NE also allows the 1 conversion (56) of rules-based networks into weight-based networks. In weight-based 2 networks, users can manually assign expression values to each node or import them from 3 online databases that are readily available in NE. The databases include Metabolic gEne 4 Rapid Visualizer (MERAV) (24), Human Proteome Atlas (HPA) (57), The Cancer Genome 5 Atlas (TCGA) (58) through Firebrowse (59), and The Genotype-Tissue Expression project 6 (GTEx) (22). The expression values can then be employed to calculate the basal level 7 expression for each node in a weight-based network using an in-built feature in NE. This   Therapeutics Editor -Developing therapeutic screens towards identification of 7 novel drug targets, drug repurposing, and personalized therapeutics 8 TISON's Therapeutics Editor (TE) assists in undertaking a therapeutic evaluation of 9 biomolecular networks developed using the NE. Users can create "therapies" by 10 employing information on drugs and their targets. Each therapy consists of a single or a 11 combination of drugs and each drug may target one or more network nodes or edges.

12
Drug action can involve: (i) the enhancement or suppression of node activity (i.e. gain of 13 function, "knock-up" or loss of function "knockdown"), (ii) node removal ("knock-out"), (iii) 14 node addition ("knock-in"), or (iv) regulation of target node expression by modifying edge 15 interactions. For rules-based networks, node knock-up or knockdown is implemented by 16 assigning a fixed user-defined value to the node thereby suppressing its upstream 17 regulation. Wherein, in the case of rules-based node knock-out, the node is deleted from 18 the network, and its rule is removed; whereas for the knock-in case, the new node along 19 with its associated rule is added into the network by updating network rewiring. Similarly, 20 in the case of weight-based networks, knock-up or knockdown is implemented by fixing 21 the node expression value along with suppression of its upstream regulation and keeping 22 its basal value at '0'; knock-out is performed by updating the node's basal value and fixed 23 node value to '0', and deleting all of its outgoing as well as incoming interactions; weight-24 based node knock-in is implemented by defining a new node, assigning its upstream and 25 downstream regulation and adding its basal value. Basal value in TE can be added using 26 two ways, users can either directly assign a node's basal value at the time of node 27 creation or, it can be calculated using expression data provided by the user towards 28 developing a personalized cancer network model. Lastly, users can also alter, knock-in, 29 or knock-out the edge weights between source and target nodes in weight-based 30 networks by updating or adding their interaction value. For such therapy-targeted nodes, 31 specific scores can also be imported from the Drug-gene Interaction Database (DGIdb) 32 (63), for ease in implementation.

33
Towards creating targeted therapy, TE users can proceed in two modalities i.e. 14 To validate and demonstrate TE's functionality, we recreated two published case 15 studies. In the first case study, we replicated the cellular response to DNA damage using   can then be used in a "decision box" for choosing between different cell fates, under 10 specific user-defined conditions. To further facilitate the users, several ready-to-use cell 11 fates including mitosis, quiescence, cell death, migration, and differentiation have also 12 been provided in CCE. Alongside this, an intuitive feature for the consumption and 13 production of environmental biomolecules has been provided (see Supplementary   14 Material, Section 2.4 for further details). Lastly, during the designing of cell circuits, users 15 can simulate their circuits within a single time step to debug logical errors within their 16 circuits. 17 To validate the functionality of CCE, we have reconstructed a literature-based case     Multi-scale modeling in systems biology involves developing integrative models of genes, 30 transcripts, proteins, cells, and investigating their regulatory dynamics over diverse 31 spatiotemporal scales. Simulations of such models can be used to decode the 32 biomolecular foundations of emergent system-level properties as well as identifying novel 1 therapeutic targets (79,80). Here, we propose "Theatre for in silico Systems Oncology" 2 (TISON), which is a web-based next-generation platform for constructing multi-scale 3 cancer systems biology models. The software embodies eight scale-specific editors 4 featuring enriched graphical user interfaces (GUIs) for intuitive model development.

5
The multi-scale model building process begins with TISON's Networks Editor (NE) 6 wherein users can construct and analyze biomolecular regulatory networks. Users can 7 also plugin expression data from online databases including Metabolic gEne Rapid    This work also exemplifies the reconstruction of several published case studies.

30
The data generated from these case studies helped decode mechanisms driving the     performing therapeutic evaluations. Network analyses for each therapy can employ DA, 16 PA, and ODE (as described in NE, above). Additionally, Plotly v1.39.2 (87-89) was used 17 to construct and plot stack and bar graphs for comparing analyses results as well as for 18 landscape construction and visualization.    simultaneously. This is followed by environmental layer diffusion. The process is repeated 15 at each time step and the resulting data is stored. This process iterates until the final time 16 step of the simulation.

17
Data and Software Availability 18 The software, case studies, and manual are freely available at http://tison.lums.edu.pk. 19 The issues reporting and database are catered at https://github.com/BIRL/TISON/issues.  Interplay between gene expression noise and regulatory network architecture.    Chaste: An Open Source C++ Library for Computational Physiology and Biology. 16 PLoS Comput Biol. 2013;9(3).