Charlotte Mary Deane

SAbDab2 provides updated antibody structures with machine learning-ready data and standardized splits for research.

GenCDR uses LLaMa-based models to generate antibody CDRs with framework context, achieving high recovery and developable designs through alignment-driven training.

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.

AbSLang-search efficiently retrieves structurally similar antibodies from large sequence databases using a contrastive learning model for CDR RMSD prediction.

This study evaluates AlphaFold 3's ability to predict T cell antigen specificity using structural models and finds that binding scores correlate with TCR positioning over pMHC, not chemical interactions.

The paper identifies synergistic antibody interactions against the malaria protein RIPR, showing how combinations of antibodies targeting specific domains inhibit parasite invasion.

This paper introduces FlashABB, a model that predicts antibody structures faster than sequence embedding models, enabling efficient access to 3D information for antibody stability and developability prediction.

This paper introduces ProtEGNN, a model combining protein language models with geometric equivariance to improve protein property and function prediction with fewer parameters than fine-tuned PLMs.

The paper reverse-engineers a TCR to reduce cross-reactivity while maintaining potency by reverting specific residues.

ABB4-STEROIDS generates antibody conformational ensembles with state-of-the-art accuracy, outperforming existing methods in experimental benchmarks.

The paper introduces ValDX, a framework to validate and quantify uncertainty in HDX-MS data integration with protein structural ensembles, showing it outperforms conventional metrics in distinguishing correct from incorrect models.

This paper introduces Pox-AbDab, a database of antibodies against orthopoxviruses, highlighting challenges in cross-species antibody function and identifying gaps in research.

This paper improves diffusion-based molecular docking by using reinforcement learning to enforce physical validity, enhancing pose accuracy without increasing computational cost.

This study introduces a molecular dynamics model and databases to analyze CDR loop flexibility in antibodies and TCRs, enabling detailed study of their dynamics and machine learning training.

ReX infers residue-level hydrogen deuterium exchange from HDX-MS data, showing improved performance in predicting unseen data and quantifying conformational changes in proteins.

The paper designs a selective protein degradation system using bumped IMiD analogs and zinc-finger degrons to target specific proteins without affecting endogenous substrates.

The paper introduces two conditioning strategies for flow-matching models to generate bioisosteric 3D molecules without retraining, demonstrating effectiveness on drug-relevant tasks.

The paper predicts the structural flexibility of antibody and T-cell receptor CDR3 loops using a new method trained on a large dataset of conformations.

STCRpy provides a Python package for TCR structure analysis and machine learning dataset preparation, enabling tasks like complex orientation scoring and interaction profiling.

The paper introduces TNP, a computational tool for predicting nanobody developability, calibrated with clinical-stage sequences and experimental data.

LICHEN generates light-chain antibody sequences conditioned on heavy chains and experimental needs, producing valid, diverse, and functional sequences.

AntiDIF improves antibody inverse folding by increasing sequence diversity while maintaining high sequence recovery rates and structural accuracy.

The paper evaluates the need for more diverse and larger datasets to improve antibody-antigen ΔΔG prediction using Graphinity, showing current experimental data is insufficient.

MolSnapper improves structure-based drug design by conditioning diffusion models with 3D pharmacophores, generating more valid molecules with higher structural and chemical similarity.

ANARCII is a language model that numbers antigen receptor sequences without alignment, achieving high accuracy and efficiency for diverse sequences.

AF2χ predicts protein side-chain rotamer distributions using AlphaFold2, outperforming simulations and existing methods in speed and accuracy.

This paper introduces a tool that integrates novel chemical reactions into retrosynthesis tools, improving performance in 70% of cases and reducing synthesis costs by 20%.

The paper evaluates TCR structure prediction models using a large dataset, finds comparable structural diversity in alpha and beta chains, and provides over 1.5M predicted structures for further analysis.

The paper introduces cryoPARES, a deep learning method for efficient cryo-EM image alignment, enabling faster and automated structure determination with reduced manual intervention.

This study evaluates machine learning-based scoring functions by comparing them to baseline models that only learn dataset biases, revealing that the MLBSFs primarily capture biases rather than generalizable physics knowledge.