Predicted Contact and Distance Map Used in D-I-TASSER-AF2
Contact Map
Distance Map
D-I-TASSER-AF2 simulation is guided by the consensus contact map
(left figure, top 2L) and distance map (right figure) derived based on confidence scores of
AlphaFold2, AttentionPotential and DeepPotential.
In the contact and distance maps, the axes mark the residue index along the sequence.
For the contact map, each dot represents a residue pair with predicted contact,
while for the distance map a color scale represents a distance of 2-22+ angstroms.
D-I-TASSER-AF2 simulations generate a large ensemble of structural
conformations, i.e. decoys. These decoys are clustered by
SPICKER based on pairwise structure similarity to
report up to five final models from the five largest clusters. Models are
ranked in descending order of cluster size. If the simulations converge
well, it is possible to have less than 5 models generated, which is
usually an indication of good model quality.
(b)
The model confidence is quantitatified by estimated TM-score, calculated based on
significance of threading template alignments, contact satisfaction rate,
mean absolute error between distnace of model and distance of predicted map, and convergence of
D-I-TASSER-AF2 simulations. higher estimated TM-score signifies higher model confidence.
References:
1.
Wei Zheng, Yang Li, Chengxin Zhang, Robin Pearce, S. M. Mortuza, Yang Zhang. "Deep-learning contact-map guided protein structure prediction in CASP13." Proteins: Structure, Function, and Bioinformatics, 87: 1149-1164 (2019).
