Justas Dauparas

This paper designs a protein family that binds and senses small molecules using deep learning and physics-based methods, achieving nanomolar sensitivity for cortisol.

This study presents a joint optimization approach for modeling protein conformational landscapes and sequence design, showing improved stability prediction and sequence generation.

An atom-level graph neural network that denoises corrupted structures generates small-molecule and protein-ligand conformational ensembles, and its active-site pre-organization metric enriched for more active designed retroaldolases.

This paper introduces NA-MPNN, a unified deep learning model for nucleic acid inverse folding, which outperforms existing methods in RNA sequence design and protein-DNA specificity prediction.

This paper presents a computational strategy to design bifaceted protein nanomaterials with two distinct sides, enabling anisotropic functionalization and controlled structural variation.

This paper introduces LigandMPNN, a deep learning method for protein sequence design that models non-protein atomic contexts, achieving better performance than existing methods in recovering interacting residues.

The paper designs nanomolar binders for diverse small molecules using a pseudocycle-based shape complementarity method, enabling direct conversion into sensing systems.

This paper introduces ProteinMPNN, a deep learning model that improves protein expression, stability, and function by incorporating evolutionary and structural information.

The paper designs closed repeat proteins with central pockets using deep learning and validates them in E. coli, showing structural consistency and potential for small molecule binding.

This study designs soluble analogues of integral membrane proteins using deep learning, achieving high thermal stability and functional accuracy.

The paper fine-tunes AlphaFold to predict peptide binding specificity for MHC and other domains, achieving performance comparable to sequence-based methods.

A sequence-only protein language model, sampled under a structure-compatibility energy, produced designs of which 152 of 228 tested behaved as soluble monomers, some with no detectable UniRef90 match.

This paper introduces a differentiable Smith-Waterman algorithm to jointly learn multiple sequence alignments and downstream tasks, showing improvements in contact prediction and structure prediction.

The paper shows that a single attention layer can predict protein contacts as well as Potts models and pretrained Transformers, suggesting attention models can capture meaningful protein features from unlabeled data.

The paper introduces a 3-track network for accurate protein structure and interaction prediction, achieving performance close to DeepMind's CASP14 results and enabling rapid modeling of protein complexes from sequence data.

The paper introduces DeepAccNet, a deep learning framework that improves protein structure refinement by estimating accuracy and guiding Rosetta refinement.

This study evaluates how well generative sequence models capture structural and fitness patterns from protein families, finding that structural and fitness information are not necessarily correlated.