This repository holds scripts for benchmarking the structure-based domain annotation approach. Domains are predicted by aligning query proteins against a database comprised of domain instance structures and assigning the Pfam family of the top-scoring instances to the queries. The alignment are generated using the following commands:
For Foldseek:
foldseek easy-search $QueryProteins $TargetDatabasePath aln_Tb_pf_gr.tsv tmp --max-seqs 1000000 -e 0.001 \
--format-output "query,target,fident,alnlen,mismatch,gapopen,qstart,qend,tstart,tend,qlen,tlen,evalue,bits,alntmscore,lddt,prob,tcov" --greedy-best-hits
For MMseqs2:
mmseqs easy-search $QueryProteins $TargetDatabasePath aln_Tb_pf_gr_seq.tsv tmp --max-seqs 1000000 -e 0.001 \
--format-output "query,target,fident,alnlen,mismatch,gapopen,qstart,qend,tstart,tend,qlen,tlen,evalue,bits,tcov" --greedy-best-hits
The output files were parsed to remove the file extensions and prefixes and to add the header to the table of hits. Predictions are benchmarked against Pfam v35.0 and Pfam-N as gold standards.
Jupyter Notebooks used for benchmarking can be found in the scripts directory. Information about the files used by workflow can be found in the README.md file inside the scripts directory.
Before running the Jupyter notebooks, make sure to download the necessary raw files. You can do so with the following commands:
wget https://zenodo.org/records/10246816/files/BenchmarkingPfamSDB.tar.gz
tar -xzf BenchmarkingPfamSDB.tar.gz