PhD Candidate · Biophysics · Harvard University

Benjamin Fry

Designing protein small-molecule interactions. Interested in geometric deep learning, structural biology, and drug discovery.

Contact Google Scholar GitHub

About

I am a PhD candidate in the Polizzi Lab at Dana-Farber Cancer Institute & Harvard Medical School, advised by Prof. Nicholas Polizzi. My research focuses on developing new algorithms for the de novo design of proteins that interact with small-molecules with applications in drug delivery and biosensing. I led the development of LASErMPNN: a graph neural network optimized for the structure-conditioned design of proteins that bind small molecule ligands. While developing LASErMPNN, we discovered that iterating rounds of LASErMPNN sequence design with protein-ligand co-structure prediction in a protocol we call NISE (Neural Iterative Selection and Expansion) enables the design of drug-binding proteins with unprecedented success rates and affinities when combined with the proper filters and ranking criteria. My current research is focused on refining NISE and using it to design binders for a wide range of small molecule ligands. NISE has produced binders to over a dozen diverse drug-like targets experimentally verified so far! After graduation, I plan to transition from academia to industry with the goal of continuing my work at the interface of machine learning, structural biology, and therapeutic design. I am passionate about contributing to open-source software and have contributed to several open-source packages including my own and others widely used by the structural biology community (e.g. Boltz)

Designed Proteins

Interactive visualization of crystal structures of some de novo designed drug-binding proteins generated with LASErMPNN and NISE. See our article in Nature for more info on these proteins!

Apixaban-binding Protein Exemplar (APEX)

Crystallographic structure of (1.9 Å) APEX protein bound to its target ligand Apixaban.
Kd = 80 ± 40 picomolar
To our knowledge at the time of publication, APEX is the highest affinity drug-binding protein ever designed by computation alone.

Exatecan-protein Interaction Construct (EPIC)

Crystallographic structure (2.21 Å) of the EPIC protein bound to its target ligand Exatecan.
Kd = 8 ± 1.6 nanomolar

Publications

Curriculum Vitae

↓ Download PDF

Education

PhD in Biophysics Harvard University · Cambridge, MA Advisor: Prof. Nicholas Polizzi · Thesis: "Design and Discovery of Novel Protein-Ligand Interactions Leveraging Synthetic Data and Deep Learning Methods" (Working Title)
BA in Biophysics Johns Hopkins University · Baltimore, MD Minor in Computer Science

Research Experience

Graduate Research Assistant Polizzi Lab · Dana-Farber Cancer Institute / Harvard Medical School
  • Developed LASErMPNN, an equivariant graph neural network for ligand-conditioned protein sequence design conditioned on 3D structure.
  • Identified NISE: a binder optimization protocol (Neural Iterative Selection and Expansion) that leverages protein-ligand co-structure prediction neural networks (RoseTTAFold-All Atom, Boltz, AlphaFold3) in conjunction with LASErMPNN to jointly optimize a ligand-binding protein's sequence and structure.
  • Implemented a structural bioinformatics informed filtering and ranking pipeline to select LASErMPNN designs achieving unprecedented experimentally determined success rates and affinities over a range of small molecule targets.
Rosetta Commons Research Experience for Undergraduates (REU) Student Kulp Lab · Wistar Institute
  • Implemented "Cloaking with Glycans" protocol to scan protein structures for N-linked glycosylation sites away from sites of interest to guide neutralizing antibody development.

Awards & Fellowships

Biophysics Prize Fellowship Marci and Martin Karplus Family Foundation
Mini Grant Recipient Rosetta Commons
Graduate Research Fellowship (GRFP) National Science Foundation

Teaching

Teaching Fellow — LIFESCI 1a Harvard University The LS1a course merges chemistry and molecular/cellular biology, focusing on applying these principles to human diseases. Teaching duties included leading a weekly chalk-talk style review of lecture material and guiding students through lab exercises (e.g. PCR, SDS-PAGE, and protein structure analysis with PyMOL).

Skills

Programming
Python, bash, C/C++
Libraries
PyTorch, ProDy, NumPy, Pandas, Matplotlib, Seaborn, Boltz-1/2
Tools
Git, SLURM, Docker, Singularity

Contact

Happy to discuss research, collaborations, or opportunities.