ATTRACT [1] is a protein - protein/RNA/DNA docking software. It uses a coarse-grained representation and a knowledge-based soft Lennard-Jones potential. The ligand is moved by gradient-based minimisation from many initial positions around the receptor. Several conformations of the ligand can be used simultaneously.
ssRN'ATTRACT [2] performs fragment-based docking of ssRNA, by docking multiple conformations for each trinucleotide (3-nt) in the RNA sequence, then assembling the docked fragments into a continuous RNA chain.
deepATTRACT [3] was created to tackle the problem of ssRNA docking into deep cavities of the protein, where 3-nt could not enter from their external initial position by simple gradient-based minimisation.
deepATTRACT uses dense grid points as starting positions for the ligand, and applies hierarchical filters to retain a reasonnable number of suitable starting positions for gradient descent minimisation in ATTRACT ff.
At first, deepATTRACT selects points of the grid surrounded by a sufficient volume to accomodate a 3-nt. At each of those points are placed each of few idealised 3-nt conformations with 128 different orientations. The combinations [point * orientation * conformation] that are free of atomic clashes are retained, and 3-nt conformers from the 3-nt library close to the corresponding ideal conformation are placed at the corresponding point * orientation combination. Those points, associated with all their clash-free orientations and conformers, are used as starting 3-nt positions for the docking with ATTRACT.
In principle, any grid of points can be used, but we only tested the usage of POCASA. The pocket finder POCASA [4] identifies possible binding pockets in/on a protein, and returns a grid of pocket points.
-
Use POCASA (probe2A, 1A spacing grid), select all pockets Merge all relevant pockets into pockets.pdb
-
Select points with at least n neighboring points within x Angstrom This ensure a minimum volume around the point to accomodate a 3-nt. Recommended x is in range [7-10], depending if you dock purines or pyrimidines. Recommended n is in range [500-1000].
./find_neighbored_points.py pockets.pdb $x $n > clusters-$xA-n$n
./deepATTRACT.sh
_ Create starting positions by applying 128 rotations at each point _ Cluster conformers at 3A, list their centers in clust3Ar.list _ Score each conformer of clust3Ar.list at each position _ Select poses (position + conformer) having a score < 1000 _ Distribute to each starting position the conformers in the same clust3Ar as the well-scored conformer _ Score, retain the e7 best-scored poses _ minimize vmax=100 _ keep top e6
[redaction ongoing] Use 1.3A cutoff for overlapping fragments assemble 5 to 8 frag > ~ 1-2.e6 chains
[1] Martin Zacharias. Proteins 2003 [2] I. Chauvot de Beauchene, S.J. de Vries, M. Zacharias. PLoS comput 2016 & NAR 2016 [3] C. Singhal, Y. Ponty, I. Chauvot de Beauchene. RECOMB 2018 https://hal.inria.fr/hal-01925083/document [4] Yu, Zhou, Tanaka, Yao. Bioinformatics 2010