Fig. 2 | Nature Communications

Fig. 2

From: A deep learning framework to predict binding preference of RNA constituents on protein surface

Fig. 2

Data statistics and performance of NucleicNet. a Statistics and hierarchy of a 7-class classification. Ratio of data available for each class is shown in the second line of each box. Each grid point is first classified under 4 coarse labels – non-site, phosphate, ribose, and base – and then 4 fine base labels – Adenine(A)/Guanine(G)/Cytosine(C)/Uracil(U). Color code of each constituent is shown in as a square. b Benchmark of NucleicNet to distinguish sites from non-sites. All methods13,14,15,16 listed from the recent review paper11 were compared using Matthew’s Correlation Coefficient (MCC) in terms of the two cutoffs, in angstrom, as indicated in title. c Benchmark of NucleicNet to distinguish among the six RNA constituents and non-sites in 3-fold cross validation of the protein-RNA complex structures in PDB. A histogram of macro-averaged accuracy is provided. Baseline accuracy (0.23) referring to a random 7-class predictor is indicated with the dash line

Back to article page