My name is Brendan Hall and I'm a fifth year PhD student in the Biophysics program at UCSF in the lab of Brian Shoichet (and previously Mike Keiser). In April I'll be starting as a scientist at Montara Therapeutics.

I went to Williams College where I got degrees in Physics and Math and modeled the kinetics of translation elongation and codon dependent ribosomal stalling with Daniel Aalberts.

At UCSF I work in structure based drug discovery and develop methods to more efficiently traverse chemical space with molecular docking.

In my free time I enjoy playing pickup volleyball and exploring the bars and restaurants in San Francisco.

At a high level I'm interested in cheminformatic and machine learning approaches to drug discovery with an emphasis on virtual screening and molecular docking.

While most of my research has focused on improving how we sample molecules from chemical space (Retrieval Augmented Docking), I have a longer-term interest in combining these advancements with improved scoring functions. While virtual screening still relies heavily on physics-based scoring, I believe the growing wealth of biophysical data will enable models that more accurately and quickly predict binding affinity. Combined with improved chemical space sampling, these models will enable the rapid and accurate screening of chemical libraries of tens of billions of molecules and beyond.

I am also interested in moving beyond affinity prediction to study how computational methods can better predict ligand bias, cell and blood-brain barrier permeability, and broader ADME properties, and how these predictions can be more effectively integrated into virtual screening workflows.

You can reach me by email at: brendan.hall@ucsf.edu

You will soon be able to find me on linkedin when I decide to spend a quarter of the time from this site on it.

Retrieval Augmented Docking (RAD)

Molecular docking is a method to computationally predict how molecules bind to proteins. Docking the tens to hundreds of billions of commercially available molecules takes significant computational resources and considerable time. We found that organizing chemical libraries into Hierarchical Navigable Small World (HNSW) graphs allows us to recover a majority of the best molecules while docking only a fraction of the total library. We demonstrated this method retrospectively, screening a library of 100 million molecules against 2 receptors with 2 different scoring functions.

For more info see the website, the paper, or the github repo.

lsd.docking.org

While molecular docking campaigns of hundreds of millions to billions of molecules are performed, the results are rarely shared in full. We developed a website providing access to poses, scores, and in vitro results for molecular docking campaigns against 11 targets, with 6.3 billion molecules docked and 3729 compounds experimentally tested. We used the new database to train machine learning models to predict docking scores and combined these models with my chemical space exploration method, Retrieval Augmented Docking.

For more info see the paper or the database.

Large scale prospective evaluation of co-folding across 557 Mac1-ligand complexes and three virtual screens

While deep learning co-folding methods (Alphafold3, Boltz-2, Chai-1) can help address the challenges of predicting ligand-bound protein complexes and ranking them, their evaluation has been hampered by data leakage and insufficiently large test sets. We tested the ability of co-folding methods to predict the structures of 557 ligands bound to the SARS-CoV-2 macrodomain. We also assessed whether co-folding methods could rescore molecular docking hit-lists to distinguish true ligands from non-binders among hundreds of molecules tested against AmpC β-lactamase, and the dopamine D4 and the σ2 receptors.

For more info see the preprint.

Ribosome Profiling Data

Ribosome profiling is a method to identify ribosome pause sites within the transcriptome. In E. Coli, the method requires an enzyme that demonstrates significant sequence bias during RNA cleavage, hampering the ability to precisely determine the position of ribosomes. We used 3' count assignment, pooling, and a debiasing method to more accurately predict the location of ribosomes. We used this improved data to show that relative codon stalling rates are not strongly correlated with relative tRNA concentrations as frequently suggested.

Preprint coming soon

This Website

One day I saw that Google was offering .phd domains and figured it was a sign to make my own website (instead of a linkedin). I wanted something interactive that showcased my interests in small molecule drug discovery. I saw a cool project called 3Dmol.js which provided nice molecular visualizations and after modifying it slightly, this is what I ended up with. Hope you like it (and that it mostly worked while you were here) :)