refineD is a new paradigm in protein structure refinement that uses machine learning based restrained relaxation. Given a starting structure, it applies deep discriminative ensemble classifiers and predicts multi-resolution probabilistic restraints to be integrated into atomistic Rosetta Energy Function during restrained relaxation. Two modes of structure refinements are available: i) a conservative mode that applies cumulative restraints for consistent refinement; and ii) an adventurous mode that applies non-cumulative restraints aimed at producing substantial degree of refinement. In either mode, five refined structures are finally produced by scoring a pool of generated refined structures using weighted combination of the deep discriminative classifiers.



refineD on-line (example output)
Example
Clear

Example Reset

 


Supplementary Materials


• training dataset:  3DRobot_set_lite_GDT_HA

• benchmark dataset:  startnative

• effects of multi-resolution probabilistic restraints:  0.5Å1.0Å2.0Å4.0Å

• refineD with cumulative (C) and non-cumulative (NC) restraints:  refineD-CrefineD-NC

• unrestrained and restrained FastRelax as controls:  FastRelaxFastRelax-0.5ÅFastRelax-1.0ÅFastRelax-2.0ÅFastRelax-4.0Å

• FG-MD and ModRefiner with restraint strength 0 and 100:  FG-MDModRefiner-0ModRefiner-100



References:

Bhattacharya, D. (2019) refineD: Improved protein structure refinement using machine learning based restrained relaxation. Bioinformatics, in press. [PDF]