Protein-Protein Docking with Spiking RAM Usage

[sl432]

Hello everyone! I’m currently working on a docking project involving two proteins, each about 300 amino acids long, using mpiexec for MPI parallel tasks with -nstruct 10000. After adding the constraints, the computation time skyrocketed from 20 hours to over 100 hours, and the RAM usage spiked significantly 😅. I’m wondering if anyone has tips for reducing memory usage and improving computation speed?

The cmd I am using is:
mpiexec /path/to/docking_protocol.mpi.linuxgccrelease -s proAB_ppked.pdb -nstruct 10000 -use_input_sc -spin -dock_pert 5 20 -partners BA_DC -ex1 -ex2aro -extra_res_cen ALE.cen.params THA.cen.params -extra_res_ra ALE.fa.params THA.fa.params -constraint:cst_file /path/to/cstfile -constains:cst_fa_file /path/to/cstfile/ -constraints:cst_weight 10 -constraints:cst_fa_weight 10 -out:file:scorefile score.sc -score:docking_interface_score 1 -overwrite > ppdockinglog.txt

cst file be like:
AtomPair N26 298B N10 660D HARMONIC 9.6 2.2
AtomPair N2 298B N10 660D HARMONIC 7.2 3.1

Here’s what I’m considering so far:

1. Is mpiexec the best approach for this? I’ve been using mpiexec to ensure flexibility in allocating cores, but I noticed most people seem to directly use -np to specify the number of cores. I’m not sure if switching to -np would make a difference.

2. Could the memory issue be due to computational saturation? I tested -nstruct 5000 structures with constraints, and the interface scores didn’t differ much from the results with -nstruct 10000. Could this be why the memory usage exploded?

I’d appreciate any insights or suggestions!