Vikram Sundar

I am a computational structural biology and machine learning researcher with a strong background in math and physics. My research interests are in using machine learning and other computational methods to better understand important problems like protein structure and binding with pharmaceutical applications. I work as a machine learning scientist at Generate:Biomedicines and just finished my PhD in MIT's Computational and Systems Biology program as a Hertz Fellow. Previously, I worked as an AI resident at Google and finished an MPhil at the University of Cambridge on a Churchill Scholarship. My PhD research was focused on engineering the specificity of TEV protease to cleave alternative substrates. During my PhD, I developed FLIGHTED, a method using Bayesian inference to denoise high-throughput experiments. I also examined the use of protein language models and very large fitness datasets to predict fitness and design new proteins, learning a great deal about dataset design and model selection in the process. Previously, I worked in the application of machine learning to understand protein/ligand binding and to model DNA-encoded libary datasets; specifically, I developed tools to measure the ability of models to generalize and attribute, new models that proved better at generalizing to unseen proteins, and uncertainty prediction methods for molecular property models. I have also completed internships on diffusion models for de novo design of protein binders at Generate:Biomedicines and on the use of molecular dynamics with permeability modeling at Inductive Bio. Skills: • Capable of directing my own research project, identifying interesting and important questions, and generating simple solutions. • Experienced in collaborating closely with experimentalists and enabling experiment-computation closed-loop cycles. • Machine learning, including traditional approaches and modern deep learning. • Bayesian modeling and inference. • Generative modeling, including diffusion models and variational autoencoders. • Generalization, interpretability, and careful experimental design in evaluating machine learning models. • Linear algebra, analysis, and other areas of math and physics, especially as applied in machine learning. • Application of computational methods to solve important structural biology problems. • Strong verbal and written communication skills, as demonstrated by publications, posters, and presentations at conferences. • Comfortable with Python, scikit-learn, tensorflow, and pytorch.