Michael M. Bronstein

Latent generative search unlocks de novo design of untapped biomolecular interactions at scale

Reward-guided inference-time search over the Proteína-Complexa sequence-structure codesign model yields validated binders across 127 targets, kinases and a free carbohydrate, with codesigned sequences outperforming ProteinMPNN redesign.

A Unified 3D Generative Model for Synthesizable Structure-Based Drug Design

This paper introduces LDDM, a generative model for drug design that produces synthetically accessible compounds validated experimentally for five protein targets.

FragmentScope - exploring the fragment space with learned surface representations

FragmentScope uses learned protein surface representations to guide fragment placement and small molecule design, achieving accurate placement and promising lead compounds.

Democratizing Protein Language Model Training, Sharing and Collaboration

The paper introduces SaprotHub, a platform that simplifies training and sharing of protein language models, enabling non-experts to build and collaborate on customized models.

Combinatorial prediction of therapeutic perturbations using causally-inspired neural networks

PDGrapher predicts combinatorial therapeutic targets to reverse disease phenotypes using a causally inspired GNN, outperforming existing methods in cell line and genetic perturbation datasets.

Target Preference Maps: A machine learning model generalizing transferable drug-receptor interactions and guiding drug discovery

The paper introduces a machine learning model that learns spatial preference maps for drug-receptor interactions to guide drug discovery.

Targeting protein-ligand neosurfaces using a generalizable deep learning approach

This paper designs proteins that target neosurfaces from protein-ligand complexes using a deep learning approach, validated experimentally against three drug-bound protein complexes.

Exploring "dark matter" protein folds using deep learning

This paper introduces a deep learning framework called Genesis to learn protein structural patterns and demonstrates its ability to design novel protein folds with high protease resistance.

PLINDER: The protein-ligand interactions dataset and evaluation resource

PLINDER provides a large, annotated PLI dataset with diverse annotations and improved evaluation splits to better assess computational methods for drug design.

PINDER: The protein interaction dataset and evaluation resource

PINDER provides a benchmark dataset for evaluating protein-protein interaction docking models, revealing overestimation of performance on leaky test sets and showing improvements with interface-clustered training.

Decoding Surface Fingerprints for Protein-Ligand Interactions

The paper introduces a computational framework for fragment-based screening and binding affinity prediction using protein surface descriptors to identify ligands for challenging protein targets.

Fast end-to-end learning on protein surfaces

This paper introduces a deep learning framework for protein structure analysis that processes raw atomic coordinates directly, achieving state-of-the-art results in interaction site identification and protein-protein interaction prediction with faster speeds and fewer parameters.