The overarching goal of my lab is to develop novel statistical methods that combine theory and computation to extract useful information from biological information and provide tools for the whole research community.  

One million human genomes, will it make a difference?  It is widely appreciated that the availability of genome information is potentially transformative for medicine. It is less widely understood that the tools we have for working with genome data are profoundly unsatisfactory. My lab concentrates on four interrelated areas of biology that address the challenge of inference from biological information, especially across scales: three-dimensional structure and dynamics of proteins, RNA; effects of mutations and fitness especially for viral evolution, immunogenomics and biotherapeutic molecular design. We have adapted and developed deep neural methods with a particular interest in unsupervised generative modeling, such as those used in natural language processing, but modified for biological sequences. We are applying and testing predictions in collaborations with experimental groups (i) predicting protein pasticity and condensates, the effects of mgenetic variants especially with respect to neurodegenerative diease and somatic mutations (ii) designing focused libraries of high affinity antibodies that bypass the need for expensive rounds of selection or labor-intensive specificity assays (iii) design and prediction of de novo proteins with specific functions. (iv) forecasting viral evolution for vaccine design and predicting cell-type specific interactions between virus and host. Across these diverse applciations we develop novel machine learning methods and statistical inference approaches to accelerate biological discovery.