Figure 1.

Flow diagram that illustrates the procedures used by THoR. A starting alignment A_n is broken into its constituent sequences and each of these is compared against the nonredundant database of choice (D_nr) using PSI-BLAST and a user-specified number of iterations. The full-length sequence for each of the high-scoring pairs (HSP-ψ) from the PSI-BLAST results are deposited in a temporary FASTA database of homologs (D_H) in a nonredundant fashion. When this database is complete, a hidden Markov model (HMM) of the alignment A_n is generated and compared with D_H using hmmsearch. Subsequently, the PSI-BLAST HSPs (HSP-ψ) and HMMER HSPs (HSP-H) are compared. They are considered equivalent if the HSP-ψ and HSP-H overlap significantly, overlap ≥ n (see text). During iterations i intermediate alignments (A_n,i) and HMMs (HMM_n,i) are produced. The latter are used to re-search the database (D_H) in order to accumulate and append additional sequences to the alignment. This iterative step is repeated until no new entries are added to the alignment when compared with the previous search (A_n,i-1). At this stage results are written to the THoR output file and the final alignment (A_n,i+1) is produced.