Patrick Bryant

The paper validates and analyzes 12,000 AI-designed CAR-T binders from a global competition, identifying successful designs and factors improving their performance.

This paper introduces EvoBind-multimer, a deep learning framework for designing molecular glues from protein sequences, demonstrating its ability to recruit E3 ligases to oncoproteins and showing context-dependent functionality.

The paper presents SHE, a phylogeny of eukaryotic proteomes based on structural comparisons, revealing a bipartite model of evolution and a framework for model organism selection.

RareFoldGPCR designs peptide agonists using noncanonical amino acids to modulate GPCRs, validated experimentally for functional activity.

This paper designs cyclic peptide agonists for GCGR and GLP1R using EvoBind, achieving dual activation with EC50 values comparable to natural hormones.

RareFold predicts and designs protein structures with noncanonical amino acids, achieving low-micromolar affinities and no immune activation in organoid models.

This paper introduces EvoBind, an AI method to design cyclic peptides that inhibit HIV-1 membrane fusion, demonstrating potent inhibition of two HIV-1 strains with no toxicity.

The paper predicts structures of human-pathogen protein interactions using AlphaFold, finding 30 high-quality models and confirming a novel heterotetrameric complex with potential immune evasion implications.

This study introduces Cfold, a neural network trained to predict alternative protein conformations, achieving high accuracy for 52% of evaluated nonredundant conformations.

The paper improves protein complex prediction by denoising the MSA profile using gradient descent through AlphaFold-multimer, increasing accuracy on difficult targets.

The paper presents a method to design linear and cyclic peptide binders of varying lengths using only a protein target sequence, achieving nM affinity and avoiding adversarial designs.

This paper introduces Umol, an AI system that predicts protein-ligand complex structures from sequence and ligand SMILES, outperforming existing methods at high accuracy.

This paper accelerates AlphaFold2-based PPI network prediction by 40x and reduces disk space by 4000x for 1000 proteins using a new protocol.

The paper uses AlphaFold2 to predict structures for 65,484 human protein interactions, identifying high confidence models and disease-related mutations.

The paper predicts structures of large protein complexes using AlphaFold and Monte Carlo tree search, achieving accurate assemblies for some symmetric complexes.

EvoBind uses AlphaFold to design peptide binders towards a protein interface using only sequence information, showing it can distinguish binders from non-binders but not affinity differences.

AlphaFold2, fed concatenated chains with optimized organism-paired alignments and pDockQ-ranked seeds, gives acceptable models for 63% of heterodimers; the same predicted DockQ score flags interacting pairs.

This paper shows that DNA methylation changes with age vary across tissues and life stages, and that non-linear relationships outperform linear epigenetic clocks in age prediction.