Using computer models to predict drug resistance in colon cancer
In the subproject “Using computer models to predict drug resistance in colon cancer” we will develop a deeper understanding of colon cancer networks and convert them into computer models with which we will be better capable to predict response to treatment. The combination of computational, experimental and clinical testing will be explored to understand drug resistance mechanisms, further paving the way towards personalised treatment of colon cancer.
About the project
Background COLOSYS project
In the COLOSYS project we set out to develop a deeper understanding of colon cancer networks and have converted them into computer models with we now use to predict response to treatment. In our approach we first gathered data on colon tumors that have been collected in international efforts and used these data to find genes that show frequent abnormal behavior in colon cancer. Such genes are believed to be ‘driving’ the disease, they play an important role in response to therapy and most importantly, these genes help us to focus on the most important parts of the colon molecular network. Next, have gathered published knowledge on molecular networks in colon cancer in a repository of data and knowledge, and we have combined this knowledge with the driver genes to construct computer models of the molecular networks in colon cancer. We have further refined these models with data collected on colon cancer cell lines (tumor cells grown in the laboratory), and are now using these models to predict response to therapy and test these predictions on cell lines and on cultures derived from patients (organoids).
The state-of-the-art of our logical modelling platform Construction of logical models
NTNU has constructed a fully automated pipeline that builds cell-type or tumour-specific models from causal interaction statements (Atopo). The resulting general topology is configured to represent specific cell types using biomarkers generated from relevant omics data (Aomics). This configuration step uses a genetic algorithm to optimise the model’s logical rules such that its steady state maximally represents the activities of the biomarkers (see below).
Configuration of logical models
Conversion of general logical models (Topology, generated with Atopo) to specific models (Logical model, generated with Gitsbe) through the use of a genetic algorithm that iteratively mutates and tests model stable state fitness against biomarker baseline (steady state) observations (Steady states) until a final model is obtained that maximally represents observed biomarkers. These models are then ready to be used for testing the effects of drug perturbations.
This deliverable is due at the end of the project. NTNU, together with members of the IMEx consortium (http://www.imexconsortium.org) is developing a checklist for the annotation of causal interactions, tentatively named MI2CAST (https://github.com/vtoure/MI2CAST) statements. This document outlines 4 rules for the metadata that ideally should be provided when annotating a causal relationship.
An initial cancer omics data repository for use in testing the different bioinformatics approaches for developing and testing models that explain cancer cell regulatory mechanisms has been assembled and is available as a table in the COLOSYS domain of the FAIRDOMhub.
NTNU has assembled a base set of causal interaction statements of the type A –> B (A up-regulates B) and C –| D (C down-regulates D) for the automatic building of logical model topologies that represent regulatory networks of specific cell types including relevant drug targets. This resource is being extended further with causal statements extracted from Reactome and other pathway databases. NTNU is working with the IntAct group of EMBL-EBI to make causal statements widely available through an upgrade of the PSICQUIC webservice.
D2.1 Comprehensive map of molecular interactions depicting the known mechanisms of driver gene action, available as NaviCell environment.
PDF document, completed June 2017.
NTNU and Curie have produced several curated high quality models for simulations of cancer cell behavior under perturbations. These models have been published already (the AGS model, Flobak et al. 2015, https://www.ncbi.nlm.nih.gov/pubmed/26317215) or will be published soon (CASCADE models), and they are listed below.
Graphical representation of the initial CASCADE model. The model contains 144 nodes and 366 edges. Edges are directed and signed. Activating edges have a green color, while inhibiting edges are colored red. Rectangular dark blue nodes represent the two output nodes. Nodes organized in a cycle represent the preliminary modules. Barbara Niederdorfer et al. Manuscript in preparation. Contact: firstname.lastname@example.org
The CASCADE 1.0 model was constructed using a bottom-up approach, extending on the AGS model and adding additional nodes and pathways that were known actors in various types of cancer, and included actionable targets that could be used in drug simulations.
Graph of the CASCADE 2.0 model. There is no distinction between the representation of genes and proteins in the model, with both being represented as cyclical, light blue colored nodes. Rectangular dark blue nodes represent the three output nodes (Proliferation/Growth, Apoptosis and Metastasis). Edges are directed and signed. Activating edges have a green color, while inhibiting edges are colored red. Nodes organized in a cycle represent the 25 modules of the model. Note the additional blue output node, indicating metastasis. Eirini Tsirvouli et al. Manuscript in preparation. Contact: email@example.com
The CASCADE 2.0 model was constructed using a middle-out approach, starting with an integrated omics data analysis produced on tumor material from patients of the TCGA-COAD cohort (Consensus Molecular Subtypes (CMS) of colorectal cancer; Guinney et al. 2015, Nat Med 21:1350–1356), and using the CMS relevant genes as seed genes to extend the CASCADE 1.0 model.
D2.3 Characterisation of all genetic interactions of the logical model with respect to all model phenotypes (CO)
A pdf document describing this deliverable was made available to the consortium partners August 2018.
This deliverable is due at the end of the project. NKI and NTNU will compare synergy predictions from logical models of colorectal cancer cells with in vitro collected drug synergy screening data and test promising drug combinations in cell lines and share results with Charité for testing in CRC organoids.
D3.1 Experimental models that are i) characterised for intrinsic resistance and ii) generated and characterised for acquired resistance (CO)
– in process of being shared among consortium partners –
– in process of being shared among consortium partners –
D3.3 Boolean models that capture resistance mechanisms and allow design of (combination) treatments that overcome resistance
Resistance mechanisms were identified using MRA (Modular Response Analysis). Procedure published in Bosdriesz E, Prahallad A, Klinger B, Sieber A, Bosma A, Bernards R, Bluthgen N, Wessels LFA “Comparative Network Reconstruction Using Mixed Integer Programming.” Bioinformatics 2018
D4.1 Database of available cell lines, 3D cultures, PDX and retrospective samples from all sites and associated molecular data
A spreadsheet listing these data sources is made available to the consortium partners through the FAIRDOMhub.
This deliverable will be based on the results of D2.4, taking validated synergistic drug combinations for testing in CRC organoids.
Data management plan
The COLOSYS consortium makes data available through the FAIRDOMhub for a persistent storage and sharing of the produced results. All data is findable and accessible at https://fairdomhub.org/projects/81. Software tools are publicly provided in software development platforms such as GitHub. In the FAIRDOMhub, a link to these platforms is provided to enable users to access the code source and implementations. Datasets are stored, or linked, in the FAIRDOMhub, from experimental datasets to computational models. It should be noted that the computational models are provided in different formats which allows to use them in diverse tools (Cytoscape, Navicell, GINsim, COLOGICS).
Overview of shareable results produced by NTNU
The shareable results include descriptions of new standards, software tools, experimental datasets and executable models:
Plan for release of additional data and software
This involves the Drug Synergy prediction platform COLOGICS and the CRISPRi data. Both will be made available through GitHub and Figshare, once the manuscripts describing these have been published.
Guidelines for the overview of all COLOSYS consortium data
The COLOSYS consortium has described its WP activities and various deliverables/datasets in the FAIDOMhub platform, following the ISA format. The structure is as follows: Each workpackages, defined as investigations, contains deliverables that are defined as studies in FAIRDOMhub. Meaning that for each deliverable, specific datasets or software tools are linked as being assays or models of a particular study.