PLMSearch and PLMAlign

PLMSearch and PLMAlign leverage pretrained protein language models to detect remote homologies and align homologous protein sequences using only sequence input.

Key Features:

• Remote homology detection: PLMSearch uses deep representations from a pretrained protein language model and trains a similarity prediction model with real structure similarities to detect remote homologies embedded in sequences.
• Sequence alignment of homologous pairs: PLMAlign accurately aligns identified homologous protein sequence pairs.
• Open resources: Open-source code, datasets, and containers are provided for local or cloud execution.

Scientific Applications:

• Homologous protein search: PLMSearch enables identification of distant evolutionary relationships using protein sequences alone, without requiring structural information.
• Protein sequence alignment: PLMAlign provides accurate alignment of remote homologous sequences identified by homology search methods.

Methodology:

PLMSearch extracts deep representations from a pretrained protein language model and trains a similarity prediction model using large-scale real structure similarities. PLMAlign aligns the homologous pairs identified by PLMSearch.