Logan Hallee

Protify benchmarks protein language models across tasks and evaluation settings, revealing task dependence and homology leakage in model performance.

This study investigates VPS34 activation in pulmonary arterial vascular smooth muscle cells in pulmonary hypertension, finding it drives proliferation and fibrotic remodeling, and shows SAR405 inhibition reduces disease progression.

The paper shows that protein language models can exploit taxonomic information rather than true interaction features in PPI datasets, leading to inflated performance.

The paper analyzes a crowdsourced protein design competition, showing participants designed EGFR binders with varying affinities and highlighting challenges in benchmarking and experimental design.

The paper introduces Annotation Vocabulary to enhance protein embeddings using structured ontologies, demonstrating its effectiveness through various downstream tasks and generative models.

This paper introduces cdsBERT, a protein language model trained on codon sequences, which outperforms amino acid-based models in enzyme prediction and reveals codon usage patterns in embeddings.

This paper demonstrates that large language models can predict protein-protein interactions with high accuracy using synthetic data and fine-tuned models.

This paper improves phylogenetic classification using codon usage frequencies and explores ORF identification through statistical tests.