Integrating topology and deep learning for bioinformatics

Guowei Wei

Department of Mathematics

Michigan State University, East Lansing, MI 48824, USA

                                                                           Email: wei@math.msu.edu

Biology is believed to be the last forefront of natural sciences. The exponential growth of biological data has set the stage for biological sciences to transform from qualitative, phenomenological and descriptive to quantitative, analytical and predictive. Mathematics, including machine learning, has become a driving force behind this historic transition as it did to quantum physics a century ago.  Geometric apparatuses are frequently inundated with too much structural detail to be computationally tractable, while traditional topology often incurs too much simplification of biological data to be practically useful. Persistent homology, a new branch of algebraic topology, is able to bridge the gap between geometry and topology. I will discuss the development of persistent homology based   machine learning, including deep learning, for cutting edge predictions of a vast variety of molecular and biomolecular properties, including drug toxicity, partition coefficients, protein-drug binding affinities, mutation induced protein stability changes and sequence based protein folds.