Mahakaran Sandhu

This study evaluates ML architectures for protein fitness prediction using simulated and empirical landscapes, identifying landscape ruggedness as a key determinant of accuracy.

The paper introduces a graph-based measure of protein fitness landscape ruggedness and shows it concentrates at regions where dominant constraints change.

This paper introduces LASE, a method using ancestral sequence reconstruction to train protein language models with higher accuracy and faster computation than large models.

The paper describes how two distinct enzymes evolved from a common ancestor through different mutational paths, leading to divergent molecular and phenotypic outcomes.