Lee, Byungwook, and Doheon Lee developed a similarity measure for protein domain architectures. This is based on precomputed ‘domain weights’. The ‘generate_domain_weights’ tool computes these domain weights for all InterPro annotations in UniprotKB in parallel using R and redis.
There are two Rscripts in the src folder that can be used to parse the large ‘match_complete.xml’ file as downloadable from Uniprot.
The two different approaches implemented are (1) parsing chunks of 10,000 lines of the above file in parallel or pre splitting ‘match_complete.xml’ into files of 10,000 lines and process those also in parallel. The latter approach is a little faster but requires more disk space and a *nix system with the split command.
1. Download the database of all InterProScan results for all proteins in UniprotKB. (See references for the link.)
2. Split the above database into files of max 10,000 lines each using the *nix shell command split.
3. Generate a sorted list of above split results, e.g. using l -1 split_files_fir > sorted_split_files_list.txt
4. On your cluster start a redis server (see references for details)
5. Submit a job to your compute cluster for each split file:
Rscript src/computeInterProDomainWeightsFromSplitFiles.R sorted_split_list.txt url.to.your.redis your.redis.port
Alternativly you may not split the large database but make R itself process different all the chunks each 10,000 lines long. In this case leave out above steps 2. and 3. and start for each chunk of 10,000 lines a single batch job (step 5.):
( First compute CHUNK_NO as number of lines in database divided by 10,000 and rounded up! )
for i in {1..CHUNK_NO}; do
l=`echo “(” $i – 1 “)” “*” 10000 | bc`
Rscript src/computeInterProDomainWeights.R match_complete.xml ${l} 10000 url.to.your.redis your.redis.port
done;Note, that the latter version of step 5. is slower than the one working on split files.
The above approach stores the number of annotations and the number of distinct direct neighbor domains for all annotated InterPro domains. These values are kept in the redis instance accessed by all parallel computations. After all of them have finished it is thus necessary to process this information into a single ‘domain weight table’.
The following script does just that:
Rscript src/generateDomainWeightsTable.R url.to.your.redis your.redis.port path/to/domain_weights_table.tbl
1. Lee, Byungwook, and Doheon Lee. “Protein Comparison at the Domain Architecture Level.” BMC Bioinformatics 10, no. Suppl 15 (December 3, 2009): S5.
2. http://redis.io/
3. http://cran.r-project.org/web/packages/rredis/index.html
4. InterProScan results for all proteins in UniprotKB: ftp://ftp.ebi.ac.uk/pub/databases/interpro/match_complete.xml.gz (last accessed September 2012)