(337ak) Prediction of Protein Structure to Function Properties for Therapeutic Purposes

With the advent and success of machine-learning and deep-learning models for protein structure prediction, protein function prediction was a natural extension to the applications of such models. My doctoral research was involved in understanding how the available protein engineering tools can be better adapted for the design of protein binding interfaces, while my postdoctoral research is based on machine-learning models for quantitative structure-to-function predictions. My research interests are centered around both my past and present works to accurately predict and design functional protein interfaces, specifically for therapeutic purposes.

The high effectiveness of proteins in biological systems can be attributed to their mutational histories. The most frequent type of mutation in proteins is the single amino acid substitution. Mutations can alter the physiochemical properties and functions of proteins and impact their folding and interactions. Studying the effects of protein mutations on protein functions was the focus of my doctoral research. Each of the projects in my dissertation was directed towards bridging the gap between computational protein structure and function by building a statistical understanding of various aspects of protein interactions and functions, thus contributing towards protein engineering techniques for therapeutic purposes. For each project, mutational analysis algorithms were used to focus on the effect of mutations on protein functions from a unique perspective: from characterizing the binding interfaces of therapeutic proteins to quantifying the effects of point mutations on protein binding, making an in-depth analysis of the effects of antigenic mutations on therapeutic protein interactions, and identifying the impact of viral mutations on immune responses in humans. The projects required protein engineering tools like forcefields for molecular simulations and Major Histocompatibility Complex (MHC)-peptide predictors. The unique sequences of proteins determine their binding interactions, and thus, their functions. This is an important aspect of MHC Class I and Class II proteins binding to peptides as a part of the adaptive immune response to antigenic proteins. This is the focus of my current postdoctoral research. Despite the availability of several machine-learning-based algorithms, predicting the binding of MHC molecules to peptides remains a challenge given the polymorphic nature of the Human Leukocyte Antigens (HLA) alleles encoding the MHC proteins. My current projects are based on the development of statistical models, for example, a multi-allele-specific Support Vector Machine (SVM) model, that can accurately predict the interactions between the antigenic peptides and the immune proteins. Understanding the binding of peptides to MHC molecules as a part of immune responses can contribute to areas of immunology like vaccine design and protein-based therapeutics.