SORS/WomenInBSC: Multi-scale cancer signaling network modeling using natural language processing

Mechanistic modeling of gene networks using ordinary differential equations (ODEs) is regarded as a promising approach to uncover regulatory mechanisms and identify drug targets in human diseases. We developed Pasmopy (Patient-Specific Modelling in Python), a computational framework for patient stratification based on in silico signalling dynamics. With this framework, we constructed a comprehensive mechanistic model of the ErbB receptor - c-Myc signalling network, trained on phospho-proteomics data from breast cancer cell lines. We then performed a simulation of 377 breast cancer patients using transcriptome data from The Cancer Genome Atlas (TCGA) as the model's initial value. Through this approach, we successfully predicted the key maker genes associated with poor prognosis of triple-negative breast cancer (TNBC) based on the time-course patterns of in silico signalling dynamics, using deep learning method.

In this study, we also developed a new computational tool called Text2model, which transforms the descriptions of biochemical reactions into mechanistic models. Building on Text2Model, we are currently developing natural language processing (NLP) methods to automatically construct mathematical models from the literature and public databases, and to predict gene-drug interactions.

In mechanistic modeling, parameters such as association or dissociation constants are often altered by gene mutation. However, predicting these parameters within cells remains challenging. We are working to address this by combining molecular-based simulations with network/pathway-based simulations. Specifically, we are currently investigating how EGFR mutations alter the parameters of EGFR-adaptor proteins binding and influence downstream ERK activity. I will explain more about our ongoing cancer modeling study in this seminar.