HIPPO (Histogram-based Pseudo-potential): Scoring Function for Protein-ssRNA Docking

HIPPO is tailored for fragment-based docking poses in the ATTRACT coarse-grained representation, generated with ATTRACT docking engine. This means that HIPPO requires a single protein structure against which all RNA models are scored. HIPPO was developed and tested on RNA fragments—specifically, trinucleotides. It can rank RNA chains/fragments of any length; however, it has not yet been tested for anything other than trinucleotides.

Requirements

To run HIPPO, you need an environment with:

• Python 3
• NumPy
• Pandas
• Parallel

Use hippo.yml to create a new hippo environment. Then export the path to HIPPO with HIPPO=/path/to/hippo. If you'd like to use HIPPO on the output of ATTRACT - make sure source in line 153 of scripts/hippo.sh points to a correct location.

Execution

To run HIPPO, you need to prepare input files and run bash $HIPPO/scripts/hippo.sh from a folder with those files.

There are two modes to run HIPPO (see bash $HIPPO/scripts/hippo.sh --help). One mode processes the output of fragment-based docking from ATTRACT, where all RNA poses are stored in a .dat file. An installation of ATTRACT is required for this mode. The second mode allows you to process RNA poses stored in a PDB file, which does not require an installation of ATTRACT.

To prepare a coarse-grained protein structure, a list of bound fragments and motifs, and either a list of PDB models of RNA or a .dat file with rotations/translations:

• To coarse-grain a PDB file, use $HIPPO/tools/reduce.py with a PDB file as an argument. Use the flag --rna for RNA. Please note that reduce.py cannot handle PDB files with multiple models.
• The boundfrag.list should look as follows:

  1 AAA
  2 AAC
  3 ACG
  4 CGU
  

  Any length or number of fragments is acceptable.
• The motif.list should look as follows:

  AAA
  AAC
  ACG
  CGU
  

• If you're using a PDB file to store models of RNA/fragments, make sure they are named frag${frag}r.pdb, where $frag corresponds to the number of a given fragment. Please ensure that RNA models are separated by ENDMDL and that all your RNA/fragment models are oriented with respect to a SINGLE protein structure. Otherwise, the resulting ranking will not make any sense.
• If you're using .dat files to store models of RNA/fragments, make sure they are named ${motif}-e7.dat.

Quick Theoretical Part

HIPPO works by scoring bead-bead contacts, using four distinct sets of pseudo-potentials. You can find out more in this paper.

Practically speaking, HIPPO will score a list of input RNA poses four times. It will then take the number of top-ranked poses specified by the user (poses_per_potential) from each of the four lists of scored poses, pooling these poses together and removing redundancies. Finally, a specified number of poses (sele_top) will be returned in the hippo.rank file.

Please note that raw histogram score values are not representative of any kind of energy.

Re-training HIPPO

The scripts used for building HIPPO are available at sjdv1982/histograms. They are customized for the ATTRACT docking engine and docking poses in ATTRACT CG representation and can be used to train HIPPO-like scoring function on any other docking benchmark, e.g. DNA.

---

Feel free to use HIPPO for your docking projects, and do not hesitate to open a GitHub issue if you have any questions! Happy docking:)