About

Hi, I’m Vlad

I’m a senior at MIT double majoring in Chemistry and Computer Science. My research interests revolve around computational chemistry, machine learning, and drug discovery with a focus on geometric deep learning and molecular dynamics.

I’m currently working under the supervision of Prof. Wengong Jin at the Broad Institute on machine learning methods for conformational sampling and binding affinity prediction. Previously, I was as an undergraduate researcher in Prof. Coley’s group where I developed an automated pipeline for free energy calculations in a lead optimization project.

Awards and Certifications

Projects

  • gmx-scripts: a set of scripts and an interactive notebook for setting up, running, and analyzing relative binding free energy (RBFE) calculations in GROMACS. The scripts include aligning congeneric ligands, force field parameterization for both the ligands and the protein, preparing hybrid topology and molecular structure files, and setting up all input files for running FEP calculations.
  • DiffDock-DSMBind: an SE(3) denoising score-matching model for unsupervised binding energy prediction built on the DiffDock codebase. The model learns to predict the energy $ E_\theta(X) $, calculates the gradient $ \nabla_X E_\theta(X) $ with respect to the input coordinates (representing atom forces), and uses this gradient to infer translation and rotation noise. The learned energy $ E(X) $ is hypothesized to correlate with experimental binding energy.
  • 6.S966 Symmetry and Its Applications to ML Final Project: Poster + Paper + Code: a NequIP-based neural network for binding affinity prediction that uses e3nn irreps to implement E(3)-equivariant convolutions.
  • mol-dataset and protein-dataset: a collection of methods for generating language model (LM) embeddings of protein sequences, featurizing small molecules, and creating PyTorch Geometric datasets for training GNNs.
  • protlig (GNNs meet Attention): a model that processes protein and small molecule embeddings, applies Attention Mechanism to weigh amino acid embeddings in the protein sequence, and then uses message passing to iteratively refine the small molecule embeddings for downstream prediction tasks.
  • lsh-similarity-search: an implementation of locality-sensitive hashing (LSH) for similarity searching of Morgan fingerprints in molecular databases.
  • 5.697 Computational Chemistry Final Project: a derivation of RESP/AM1-BCC charges and a torsion scan at the at the MP2/6-31G* level of theory for a small molecule; molecular dynamics simulation analysis of a protein-ligand complex.
  • gnn: a simple Message Passing Neural Network (MPNN) to predict the solvation free energies of small molecules using selected GAFF parameters as input features.

Tutorials and Walkthroughs

I also create tutorials on using tools like RDKit, PyTorch Geometric, e3nn, BioPython, and ESM3 for training graph neural networks (GNNs) in drug discovery. Check them out under the Tutorials tab!