SSAlign: Fast and Sensitive Protein Structure Search at Scale.

SSAlign enables database-scale protein structure search over 53.7 million structures, completing 1,000 queries in 3,150 s on CPUs and 2,290 s with GPU prefiltering.

• Why it matters: Structural comparison is becoming a retrieval problem as predicted structure resources grow, but exhaustive aligners are impractical and fast prefilters can miss remote or low-complexity homologs.
• What they did: SSAlign combines SaProt sequence-structure embeddings, entropy-reduced dense retrieval, an SS-score predictor, and two-stage alignment refinement, then benchmarks on Swiss-Prot, SCOPe40, and AFDB50.
• The result: On AFDB50 it returned more TM-align-supported hits than Foldseek in recall-oriented evaluations, and on SCOPe40 it increased AUC by 20.2% (family) and 33.3% (superfamily).