Anastasiya V. Kulikova

The paper introduces a method using sparse autoencoder features from InterPLM to predict neuropeptide precursors from secreted proteins across species.

The paper uses protein large language models to detect class II microcins in bacterial genomes, finding embeddings outperform sequence-based methods like BLAST in identifying known and novel microcins.

This study uses 3D CNNs to predict wild-type amino acids and consensus sequences from local structural context, finding high-confidence predictions correlate with correct outcomes.