Research

Natural proteins are linear, unidirectional biopolymers constructed by strategically adding one of the twenty possible amino acids after another such that they fold into a precise, non-clashing 3D-structure. Proteins come in various lengths, compositions, geometries, and functions. The structure of a protein offers grooves/ patches to fit/ attract small molecules (drugs, ions, etc.) or other proteins. The amino acid sequence dictates the structure and consequently the function that a protein can execute. While experiments can help us discern what the sequence, structure, and function of any protein are, predicting causal sequence-structure-function maps have largely remained elusive. We use a combination of deep-learning networks, physics-based molecular-mechanical models, and integer-optimization to predict amino acid transitions (mutations, insertions, or deletions) that will endow a protein with a desired functionality (with/ without significant structural change). We also predict protein activity given the pH, dielectric constant, polarity, and temperature of the surrounding solvent.

We modify known natural/ synthetic proteins or design from scratch proteins for - (a) enhanced/ altered catalysis, (b) precise molecular sieving of ions or solutes, (c) DNA/RNA-sequencing, (d) bio-based fuel cells, (e) biomedical applications and drug discovery, (f) structure-guided metabolic modeling, and (g) extracting specific rare-earth elements from electronic waste.

We are a computational group but we collaborate with experimental collegues for each of our projects. We also utilize tools from other established domains of quantitative science for analogous modeling of biological systems.

⇠ Enzymes (biocatalysis)

We use deep-learning-derived encoding of both enzymatic protein sequences and their shape to predict mutations to functionalize them for intended biochemistry. We identify/ build-on stable proteins and predict mutations to activate them at given pH and temperature.
Applications : biosensor design, production of precursors for biofuel and drugs

⇠ Protein pores (separations)

We use pore-containing proteins as chassis to design precise molecular separation systems. Mutations on the exterior of the pore permit incorporation in polymeric membranes while mutating inner pore walls controls ionic selectivity.
Applications : molecular separation, bio-based fuel cell, nanopore sequencing

⇠ Antibodies (immunology)

We design libraries of humanized proteins with complementary shape and electrostatics to neutralize disease-causing antigenic proteins. We also model antigenic protein mutation to understand escape from human antibodies and infectivity.
Applications : antibody design, CDR3 loop engineering

⇠ Strain design (metabolism)

We predict enzyme structures that participate in metabolic and signaling pathways. We estimate substrate binding and product leaving kinetics using protein-ligand simulations. Structure-aware metabolic models provide molecular bases to whole-cell behavior.
Applications : microbial engineering, temperature-aware metabolic models

⇠ Bio-based surfaces (materials)

We discern biochemical principles that guide designing and deploying functional protein-non-protein ensembles. Non-protein materials of interest include but are not limited to silica-derivatives, lignin, and synthetic block copolymers.
Applications : novel functional nano-materials, green chemistry, energy

⇠ Analogous biology (AI/ ML)

We pursue certain audacious projects where we first identify and employ analogous ideas from non-biological domains to throw light into open questions in biology. We design experiments for uncertainty minimization while parametrizing biological processes.
Applications : abiotic protein design, peptide therapeutics