Projects

Developing a framework for the co-folding of proteins and ligands:

We are building a novel neural framework inspired by recent advances in protein structure prediction, including OpenFold, that enables the co-folding of proteins and their ligands. By employing our approach, we anticipate a substantial enhancement in comprehending the biophysical context of protein structures, leading to new avenues for drug screening against previously unexplored targets.
 

Team members:

  • Milen Ferev (Master student, Columbia)
  • Christina Floristean (Graduate student, Columbia)
  • Lukas Jarosch (Master student, University of Heidelberg)
  • Chris Kim (Postodoctoral fellow, Columbia)
  • Itamar Shamir (previously Postdoctoral fellow, CMSA, Harvard)
  • Wojtek Treyde (incoming Graduate student, Oxford)
     

Developing a single-sequence model for protein structure prediction:

AlphaFold2’s performance demonstrates the remarkable power of deep learning in molecular problems when co-evolutionary information is available in terms of multiple sequence alignments (MSAs). However, AlphaFold2 is less performative with proteins that lack sequence homologs, and many applications require predicting 3D structures from a single sequence. Using OpenFold’s architectures, we are building a single-sequence model for protein structure prediction that de-emphasizes the co-evolutionary components of AlphaFold2. This model may provide insights into the physical principles of the folding process and is particularly useful in understanding the effects of mutations on protein stability. This project is being carried out in collaboration with Prof E. Shakhnovich.

Team members:

  • Lucas Gascon (Master student, École Polytechnique)
  • Dan Lesman (Master student, Oxford)
  • Akshay Manglik (Undergraduate student, Columbia)
  • Swati Negi (Master student, Columbia)
  • Kibum Park (Graduate student, Harvard. Shakhnovich's group)
  • Yiming Qu (Graduate student, Columbia)
     

Predicting RNA 3D structures:

The problem of predicting the 3D structure of an RNA from its primary sequence is another central challenge in biology, similar to protein folding, it is important for understanding RNA function, and RNA-based therapeutics. Using OpenFold, we are building an end-to-end pipeline for predicting RNA 3D structures. This project is being carried out in collaboration with Dr. E. Rivas.

Team members:

  •  Sachin Kadyan (Researcher, Columbia)
  • Magnus Marcin (Postdoctoral fellow, Harvard. Rivas' group)
  • Sid Sanghi (Incoming Graduate Student, UC Davis)
  • Marcell Szikszai (Postdoctoral fellow, Harvard. Rivas' group)
     

 Testing novel neural architectures for improving scalability:

Long-range dependencies are ubiquitous in biology, appearing in transcriptional networks, networks of neurons, and protein folding. A major challenge in machine learning for molecular modeling is to develop neural architectures capable of efficiently capturing long-range dependencies. We are testing novel architectures to improve AlphaFold2 scalability. This is important for predicting large molecular systems.