My main research interest lies at the intersection and the connecting of systems and structural biology. In particular, I am pursuing a bottoms-up approach to systems biology, whereby a systems-level understanding of biological processes is achieved through a molecular understanding of biological structures and their interactions. Such a molecular systems biology would result in predictive models of biological processes, derived from first principles and grounded in the rules of molecular interactions.

For this to happen two pieces must fall into place. First, we must be able to very accurately (angstrom resolution) predict the structures of biological molecules, including but not limited to proteins. Second, we must be able to translate that structural information into functional information that is both quantitative and molecular. In other words, structure must tell us something quantitative about the molecular binding partners of a biomolecule, instead of merely providing a high-level functional annotation. It is only at that level of molecular description can we hope to obtain a unified framework for describing and modeling biological systems as phenomena emergent from more fundamental interactions. While nature, ultimately, is all just physics, biology is ultimately all just structure.

My main tools for tackling these problems are the concepts and algorithms of machine learning. In my graduate work, I developed a computational method that synthesizes concepts from compressed sensing and statistical mechanics to predict the quantitative binding affinity of proteins to DNA from the structures of protein-DNA complexes. In the process I also developed a new type of energy potential, the basic building block of computational molecular analysis, that is fundamentally different from existing knowledge-based and physics-based potentials by virtue of being non-parametric, i.e. by not relying on any prior modeling assumptions. I plan to continue the development of such basic computational methods, as well as to apply these methods to simple biological subsystems that are amenable to structurally-driven analysis.