Machine learning for computational biology
We develop and apply machine learning (ML) and artificial intelligence (AI) methods for the analysis of large biomedical datasets. For example, we work on interpretable deep learning to make machine learning more useful for biomedical research.
Machine learning has emerged as a versatile approach for predicting a wide range of biological phenomena. However, its utility for biological discovery has been limited, given that machine learning methods (including the popular and powerful deep neural networks) typically provide little insight into the biological mechanisms that underlie a successful prediction.
We have a long-standing interest in making machine-learning algorithms more interpretable and useful for biological discovery and biomedical applications. Most recently, we have proposed the concept of “knowledge-primed neural networks” for interpretable deep learning on gene-regulatory networks which allows us to infer the activity of key regulatory proteins from single-cell sequencing data, including signaling pathways and protein activity states that are normally hidden to sequencing-based methods.
The power of interpretable machine learning for advancing biomedical research is further illustrated by a series of studies in which we applied machine learning to patient cohorts, for example for time series modelling of the drug response in patients with chronic lymphocytic leukemia and for the identification of drug new combinations for targeted therapy. Moreover, we successfully used machine learning to reconstruct normal and aberrant stem cell differentiation in the blood.
* shared first or shared senior authorship