Euretos Computational Disease Models
We work closely together with many pharmaceutical & biotech companies across all stages of the drug development cycle. We contribute to their research by building upon our library of molecular computational disease models. These computational models provide unique insights into which molecular perturbations cause cell type and tissue level dysfunction and predict their effect on phenotypes and diseases and, ultimately, drug efficacy and safety.
Building systems biological computational disease models
We build computational molecular disease models to understand which molecular processes are disturbed, in which cell types and tissues this disturbance occurs, and how communication between cells is affected.
To build these models we integrate genetic, expression, proteome and metabolome data. We link molecular disease associations to cellular, tissue and organismal phenotypes using the millions of biological relations in the Euretos knowledge base. In particular protein interactions, protein signalling and transcriptional networks are essential building blocks of our systems biology disease models.
Our guiding philosophy is to understand how molecular perturbations cause cellular dysfunction, tissue dysfunction, ultimately leading to the manifestation of disease. Conceptually, we view a disease as the result of one or multiple perturbations of a healthy or normal state. A perturbation may be a genetic mutation, a collection of common risk variants, or an environmental factor such as an infectious agent.
Data driven insights in disease and drug research
On the one hand our computational disease models are designed to provide useful predictions: what is the best target for asthma? Which indication can be treated with our agonist of a glutamate receptor? On the other hand we aim to give insight into the molecular mechanisms that underlie such a prediction. Which biological processes are affected when a certain protein is modulated? How does protein dysfunction cause disease? Mechanistic insights are critical to further the understanding of disease, identify biomarkers, and to guide experimental follow-up or clinical trials.
There are a number of important applications of our approach. In target identification we perform a perturbation analysis of the model of the disease of interest to identify targets. In indication expansion we use perturbation analysis to assess in each disease model how effective modulation of a target of interest will be. In toxicity and safety analysis, we use phenotypic predictions to predict potential side effects of modulating a certain target. We can also apply our approach to identify biological mechanisms responsible for differences in drug response, adverse events from phenotypic observations and molecular profiles.
Combining statistical modeling with machine & deep learning
We use a combination of statistical modelling and machine learning to build computational disease models and perform perturbation analysis. A statistical model encodes existing biological knowledge to interpret patterns in large scale multi-omics data sets and distill mechanistic insights. For instance, we use cell type deconvolution to estimate cell type proportions in bulk expression profiles. Comparing cell type proportions between samples of healthy individuals and those with disease may highlight the involvement of specific cell types.
However, there is still much about biology that is not known yet. Therefore we make extensive use of machine learning and in some cases deep learning, which allows us to find novel features in molecular data sets that are predictive of risk for disease. We use a variety of techniques to interpret relevant features and generate hypotheses about the biological mechanisms that underlie such predictions. For instance, we use in silico perturbation analysis to identify proteins that can correct the dysregulation profile associated with the disease based on machine learning models.