Improving performance¶
If InterProScan is taking a long time to run, or you just want to improve on the run time you are getting, then consider some of the following:
Review your CPU (and memory) command options¶
By default InterProScan uses 8 cpu cores on your machine. Most of the times this configuration is sufficient. However, if you have more cores available and you have more memory to support more threads, then you can change the number of cpu cores used by adding the option below to the InterProScan command line, where N is the desired number of cores
-cpu N
The value N for -cpu represents the maximum number of threads (embedded workers) InterProScan will start and run at a time.
You have to remember, the more cores you specify, the more memory InterProScan will require to run successfully. Here are some observed numbers that may act as a guide, but you may have to experiment for your own data. The input sequences were taken from UniProt
-cpu |
max memory used (GB) |
input sequence count |
input sequence size (MB) |
run time |
|---|---|---|---|---|
16 |
8 |
8,000 |
3 |
2 hrs |
16 |
12 |
16,000 |
6 |
4 hrs |
16 |
15 |
160, 000 |
56 |
12hrs |
Let’s say you have a super machine with 32 cores available and you want to use all or most of the cores. It would be recommended to specify -cpu 30, as the main InterProScan process will always use 1 core.
Each database analysis may also have options to specify how many threads to assign to it, for example, HMMER3 based analyses such as Gene3D have this option. But we dont recommend changing the default cpu values for each analysis.
Consider chunking large input files¶
If your FASTA input files contains a large number of sequences say over 160, 0000 protein sequences, then you may consider splitting your input into smaller chunks (depends on resources, but batches of 80,000 protein sequences is a suggested starting point). You can then submit the smaller input files to InterProScan and process the results afterwards.
For DNA/RNA sequences a much smaller number is suggested (e.g. 12,000 sequences). However for improved performance you could translate these using an external tool and then submit the necessary protein sequences instead, see running nucleic acid sequences for more information.
Review your command line input options¶
Do you need all the output InterProScan supplies by default? See How to run InterProScan for more details, for example you may consider options such as:
Which result data are you interested in, do you require all applications (see -appl option )?
Do you require the residue level annotation? If not, this calculation can be disabled with the -dra option.
Make use of the default lookup service, or your own local lookup service to avoid the need for calculating known results again (on by default, read more).
Running InterProScan in CLUSTER mode¶
This mode is still experimental, so I would not run in this mode in production.
You want to analysis sequences on a cluster/farm and you would like to set the number of reserved cores for each node.
Configure to analyse fewer ORFs (applies to nucleic acid sequences only)¶
For nucleic acid sequences, consider reducing the number of ORFs to analyse.