Fig. 3: Speed of SARST2 and several state-of-the-art alignment algorithms.

a Summary of the non-parallelized running time of the algorithms evaluated by IR using the Qry400 and SCOP-2.07 datasets. Most database search algorithms, like the SARSTs and BLAST, provide quality control and hit-list cutoff parameters for users to decide the amount of retrieved homologs. Such algorithms were evaluated by adjusting one cutoff at a time. For each query, the running time of an algorithm was measured under the minimal cutoff setting that allowed the algorithm to recall all family-level homologs. The right part lists the alignment speeds for algorithms that support one-step pairwise alignment. Although SARST2 was primarily designed for database search, its pairwise alignment was also efficient, e.g., around 12 times faster than the accurate structural alignment method FAST and 2 times faster than the pairwise BLAST sequence alignment program bl2seq. b Speed of database search algorithms that support parallel computation. As the number of CPUs increased, their speeds improved. Each point represents the average of 10 replicate runs using 400 query proteins (n = 400). SARST2 with 32 i9 CPUs reached an average speed of 966,000 alignments per second. A full search of the 145,000-protein SCOP-2.07 took only 0.15 s. The same task took Foldseek and BLAST 1.07 and 3.62 s, respectively. c Relative speed plot showing the degree of enhancement by parallel computing. SARST2 achieved the highest speedup, demonstrating its high scalability. The data of panels (b) and (c) are provided as a Source data file. CPU Central Processing Unit, s second.