Ultra-fast and highly sensitive protein structure alignment with segment-level representations and block-sparse optimization 1. This study introduces SPfast, a novel method for protein structure alignment that achieves remarkable speed and sensitivity improvements. SPfast accelerates search by over 2 orders of magnitude compared to traditional methods like TMalign and Foldseek, while also increasing sensitivity by 21% and 5% respectively. This makes it highly effective for large-scale structure library searches. 2. SPfast employs a unique segment-level representation of protein structures, replacing the traditional Cα-based structure representation. By using idealized key points extracted from secondary structure segments, it generates a minimal set of alignment seeds. This coarse-grained approach significantly reduces computational complexity and enhances alignment efficiency. 3. The method incorporates a block-sparse optimization technique for all-atom refinement of pairwise alignments. This allows SPfast to maintain high sensitivity while achieving ultra-fast execution times, making it suitable for high-throughput applications such as annotating uncharacterized proteins in large databases. 4. In benchmark tests, SPfast demonstrated superior performance in fold recognition of annotated SCOPe domains, achieving the highest mean sensitivity at the first false positive across all levels of the protein structure hierarchy. It also showed enhanced ability to identify multi-domain proteins and designed proteins with synthetic structures. 5. SPfast was used to conduct over 100 billion pairwise comparisons between uncharacterized proteins and a large-scale annotated structure library. This uncovered new biological insights, such as identifying novel toxin-antitoxin systems and type III secretion systems in pathogenic bacteria. These findings are supported by genomic context and high-confidence AlphaFold3 complex modeling. 6. The study highlights SPfast's potential for annotating "dark" protein clusters in the AFDB-clusters database. By mapping uncharacterized proteins to PFAM annotations, SPfast identified functional relationships that were not previously recognized, demonstrating its utility in expanding our understanding of protein function. 📜Paper: biorxiv.org/content/10.1101/… #ProteinStructure #Alignment #Bioinformatics #HighThroughput #ComputationalBiology