Integrated Mechanistic Modelling and Analysis of Large-scale Biomedical Data

Introduction

Modern cancer therapeutics target signalling processes within the cancer cells and the interaction of cancer and immune cells. A comprehensive understanding of these signalling processes is therefore essential to identify drug targets, plan clinical trials, and to select suitable drugs, drug combinations, and drug dosages for a specific patient.

Limitations of Current Models

Yet, most of the available mathematical models capture only a small number of molecular species and pathways, thereby ignoring important crosstalk and feedback loops. Furthermore, these models are usually based on experimental data for cell lines, which behave differently from complex cancer tissues.

Project Overview

In INTEGRATE, I will develop computational methods for the full process of data-driven modelling of signalling processes in cancer, ranging from model development to parameterisation all the way to uncertainty analysis.

Methodology

To this end, I will combine methods from the fields of:

1. Mathematical modelling
2. Machine learning
3. Signal processing

with established approaches in systems biology.

Model Development

The model development will employ:

• Natural language processing
• An automatic testing framework

Federated Model Inference

For federated model inference, I will develop:

• Scalable mini-batch optimisation
• Marginalisation-based uncertainty quantification

Model Refinement

To refine models, I will exploit tools from signal processing, such as blind identification of latent variables.

Application of the Approach

I will apply the developed scalable mechanistic modelling approach to integrate large-scale biomedical data for molecular phenotyping studies and clinical trials across sites. This will provide mechanistic models reconciling the available data.

Innovative Aspects

The study will, for the first time, combine mechanistic modelling and machine learning for the integrated analysis of patient-derived omics and phenotypic data. By linking these data sources, INTEGRATE will deepen our understanding of biological signal processing and provide the basis for the development of digital twins.