Kieran Didi

This study designs de novo amide ligases with a cysteine catalytic triad to selectively form amide bonds across diverse substrates, demonstrating high selectivity and applicability in synthesizing bioactive molecules.

This study designs de novo cysteine proteases with high activity and structural novelty, achieving 19% success rate and up to 3 x 10^7 rate enhancement over background reactions.

This paper designs de novo metalloproteases using RoseTTAFold Diffusion 2, achieving high activity and specificity for cleaving various substrates.

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.

FLIP2 adds seven protein fitness datasets and 16 splits for enzymes, light-sensitive proteins, and interactions; simple ridge and zero-shot baselines often rival fine-tuned protein language models.

Riboseek is a fast and sensitive sequence alignment tool for RNA and DNA that uses di-mers for alignment, outperforming BLASTN and nhmmer in speed and sensitivity.

The paper expands the AlphaFold Protein Structure Database to include proteome-scale quaternary structures, predicting 1.81M high-confidence protein complexes and revealing conserved structural clusters across domains of life.

The paper accelerates protein structure prediction using OpenFold with TensorRT and MMseqs2-GPU, achieving 131x faster inference than AlphaFold2 on specific hardware.

An all-atom diffusion model using a 14-atom residue representation and sparse U-Net architecture accelerates biomolecular design 10-fold, enabling experimentally validated DNA-binding proteins and multi-turnover cysteine hydrolases.

The paper introduces GPU-accelerated homology search with MMseqs2, achieving faster sequence similarity detection than CPU methods and improving structure prediction speeds.

AtomWorks, a modular atom-level data framework, trains structure prediction and design models; the resulting RF3 improves chirality handling and performs between AF3 and Boltz.

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

The paper introduces MISATO, a dataset combining quantum mechanics properties and molecular dynamics simulations of protein-ligand complexes to support structure-based drug discovery.

This paper introduces a protocol to create a FAIR collection of machine learning models in life sciences and medicine, aiming to improve reproducibility and reuse through community curation.