Mahakaran Sandhu

Investigating the determinants of performance in machine learning for protein fitness prediction

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

The composition of biophysical constraints generates complex, rugged regions of protein fitness landscapes

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

Leveraging ancestral sequence reconstruction for protein representation learning

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