Mohammed AlQuraishi

The paper develops a quantitative statistical mechanical model to predict protein interactions from sequences and uses it to analyze the structure and organization of human signaling networks.

A public crystallographic structure and affinity dataset for enteroviral 2A protease (925 binding events, 699 compounds), with docking, cofolding and screening baselines. Fragment fine-tuning of OpenFold3-p2 improves follow-on predictions.

This study evaluates ML and biophysical models for predicting TCR-pMHC specificity, showing ML struggles with novel peptides while biophysical methods excel but underperform on known ones, leading to a hybrid model with improved performance.

The paper explores random protein sequence space to identify physical phenotypes and shows that sequence-based models can predict experimentally observed properties.

This paper introduces Genie 3, a diffusion model that improves protein design by combining SE(3)-equivariance with hallucination, achieving state-of-the-art results in binder design and motif scaffolding.

ROCKET refines AlphaFold2 predictions against crystallographic, cryo-EM and cryo-ET data by gradient descent on MSA cluster profiles, capturing loop flips and domain shifts without retraining.

GREmLN is a foundation model that uses graph structure to improve gene embeddings in single-cell transcriptomics, outperforming baselines in cell type annotation and perturbation prediction.

The paper trains sparse autoencoders on a protein language model to identify features that represent protein sequences, revealing both generic and family-specific patterns and their potential biological roles.

The paper trains a new version of AlphaFold2, OpenFold, and evaluates its generalization ability and learning mechanisms.

This paper introduces Sequence UNET, a deep learning model for variant effect prediction that is highly scalable and computationally efficient, achieving comparable performance to existing methods on ClinVar data.

This paper introduces RGN2, a deep learning model that predicts protein structure from single sequences without MSAs, outperforming AlphaFold2 and RoseTTAFold on orphan proteins and achieving faster computation.

This paper introduces UniRep, a deep learning-based representation of protein sequences that enables accurate prediction of protein stability and function while reducing engineering costs.

This paper introduces an end-to-end differentiable model for protein structure prediction, achieving state-of-the-art accuracy on novel folds and showing promise for known folds without structural templates.

This paper introduces pNeRF, a parallelizable algorithm for converting polymer coordinates, achieving significant speedup in molecular modeling workflows.