BEST: Basic Embedding Search Tool Enhancing Discovery of Novel Enzyme.

Area of Science:

• Bioinformatics
• Computational Biology
• Structural Biology

Background:

• Protein homology identification is crucial for biological research.
• Traditional methods like sequence alignment struggle to detect remote homologs.
• There is a need for faster and more sensitive homology search tools.

Purpose of the Study:

• To develop a novel tool, Basic Embedding Search Tool (BEST), for fast and sensitive protein homology identification.
• To leverage protein language models for improved detection of remote homologs.
• To accelerate sequence encoding and vector retrieval for efficient database searching.

Main Methods:

• Utilized protein language models to generate sequence embeddings incorporating evolutionary and structural data.
• Implemented a segmented distillation pruning technique for accelerated sequence encoding.
• Developed a multi-layer acceleration structure for rapid dense vector access and retrieval.
• Conducted extensive experiments on real datasets to evaluate BEST's performance.

Main Results:

• BEST demonstrated over 20% increased sensitivity compared to existing methods, maintaining high precision and recall.
• Achieved a 4290.86-fold speedup in dense vector access and retrieval.
• Operated 23.41 times faster than PSI-BLAST and 3.92 times faster than Foldseek.
• Successfully identified homologous sequences missed by conventional homology search tools.

Conclusions:

• BEST offers a significant advancement in protein homology identification, outperforming traditional methods in both sensitivity and speed.
• The tool's ability to detect remote homologs and its efficiency make it valuable for enzyme mining and broader biological research.
• An open-access web server and publicly available code facilitate the adoption and application of BEST in the scientific community.