D-QUARK
Distance-based ab initio protein structure assembly
D-QUARK ('Distance-assisted QUARK') is a method for
ab initio (or template-free modeling, FM)
protein structure prediction.
D-QUARK is an extension of
QUARK with the fragment assembly simulations
guided by deep-learning based distance- and orientation-map predictions.
Starting from a query sequence, D-QUARK first constructs a muliple sequence alignment (MSA)
by searching through whole-genome and metagenome sequence databases using
DeepMSA2.
Next, inter-residue distance-map and dihedral-angle orientation are predicted
by DeepPotential through deep residual convolutional nueral network training,
where local fragment structures are created from the distance and orientation models
by Limited-memory Broyden–Fletcher–Goldfarb–Shanno (L-BFGS) optimizations.
Finally, full-length structural models are assembled by the replica-exchange fragment-assembly Monte Carlo
simulations which are guided by the deep-learning distance and orientation restraints.
D-QUARK participated in community-wide
CASP14
challenge as the "QUARK" group, which was ranked as the top automated server for FM targets.
(
Read more about D-QUARK)
Online server (view example output)
Dataset:
Reference:
- Chengxin Zhang, Yang Li, Yang Zhang (2021) D-QUARK: ab initio protein structure prediction guided by multiple deep learning predicted distance and orientation restraints