Neil Thomas

The paper uses language modeling to create a comprehensive representation of protein biology, enabling structure prediction and discovery of high-affinity protein interactions.

The paper identifies non-canonical FBN1 splice variants in undiagnosed Marfan syndrome families using genome sequencing and RNA testing, finding 20 candidate variants with significant enrichment in FTAAD cases.

This study combines machine learning and high-throughput screening to design improved nuclease enzymes, finding a more effective variant than standard directed evolution methods.

ESM3, a 98-billion-parameter multimodal protein language model tokenized across sequence, structure, and function, designs a bright green fluorescent protein at 58% sequence identity to known fluorescent proteins.

The paper introduces DeNovoLOEUF, a method to identify diagnostic variants in genome sequencing data, achieving 96% positive predictive value and uncovering 39 missed diagnoses in the 100,000 Genomes Project.

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.

This paper introduces SLIP, a framework for creating synthetic protein fitness landscapes to benchmark ML-guided protein design pipelines.

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.

Across five protein tasks with evolution-aware splits, self-supervised pretraining helped nearly every model on every task, yet alignment-based features still beat learned embeddings on secondary structure and contact prediction.