Methods using HHBlits as the PSSM generator

HHBlits is an efficient iterative protein sequence similarity search algorithm that aligns sequences according to their hidden Markov models [68]. Since the MSA conservation profile of HHBlits can be readily transformed into a PSSM, it is increasingly utilized by recent SSP methods as a PSSM generator. The HHBlits-based SSP algorithms applied in this study fall into two types. The NetSurfP-2 [10] used only HHBlits to generate its SSP feature set, while the others used both the PSSMs generated by HHBlits and PSI-BLAST as predictive features, including MUFOLD-SS [9], Porter5 [11], and Spider3 [8]. As listed in S8 Table, we also unified the version and settings of PSI-BLAST used in the pipelines of these programs. Additionally, the version of HHBlits was fixed to be 3.3.0. These HHBlits-based SSP algorithms were significantly more accurate than the PSI-BLAST-based ones. Assessed with the TS115 and CASP12/13 independent datasets, both Q3 and Q8 of the former were >3% higher than the latter (see S5 Table).