Dunbrack library takes a very long time to load with custom build

[Anonymous]

Hi,

I built PyRosetta from scratch with the latest release of Rosetta (2018.19).

Below are the cofigurations of my system:

OS details: CentOS Linux 6.7 64 bit cluster 

Python version 3.6 64 bit

I was successfully able to build the PyRosetta source. However, when I try to run a custom protocol, the Dunbrack library takes a very long time to load. Below is the output after the library loads:

**Dunbrack 2010 library took 618.78 seconds to load from ASCII**

I used the instructions from README within source/src/python/PyRosetta/README to do the build. 

I want to use my custom protocol in a high-throughput manner and in parallel and this loading of library is limiting the development. Is there a way to link the files and databases statically to the PyRosetta build? Or make this efficient in some way.

Thank you.

[Anonymous]

The Dunbrack library should only be loaded from ASCII the first time PyRosetta runs. Subsequent runs should load it from a cached binary file, which should take much less time. (Typically on the order of a second or so.) This cached binary should be automatically generated and written to the database directory the after the library is loaded from ASCII the first time.

If it isn’t substantially faster on subsequent runs, one issue might be if the user running PyRosetta doesn’t have write permissions in the (Py)Rosetta database directory. In that case, PyRosetta can’t write the cached binary file, and thus has to re-load from ASCII each time. Temporarily adding write permissions (at least until you’ve done a quick scoring run) should fix things.

[Anonymous]

Hi Rocco,

Thank you very much. I gave the write permissions explicitely and it seems to be caching the binary files for subsequent use.

[Author]

Posts