Fig 1.
Schematic illustration of PEMPNI and novel energy features.
(A) Flowchart of our computational framework. (B) Geometric partition-based energy features.
Fig 2.
Comparison of binding affinity changes induced by MPDs and MPRs.
(A) Overall comparison of MPDs and MPRs. (B) Comparison of MPDs and MPRs based on residue types. The numbers of mutations are shown in the columns. (C) Distribution of MPDs and MPRs in different geometric locations. (D) Comparison of MPDs and MPRs based on geometric locations. (E) Distances between different mutation groups. (F) Comparison of MPDs and MPRs interacting with major or minor grooves. (G-I) Comparison of MPDs and MPRs based on their major binding modes (i.e., the major interacting nucleotides (G), the major interacting subunits of nucleotides (H), and the major interacting contacts between the target residue and nucleotides (I)). For example, the major interacting nucleotides represent that the atomic contacts between a specific type of nucleotides and the target residue account for the greatest proportion among all contacts involving this residue. *** P<0.001, ** 0.001≤ P<0.01, and * 0.01≤ P<0.05.
Fig 3.
PCC values of individual and integrative feature groups for LOCOV.
(A) PCC values of energy feature groups for MPD276. (B) PCC values of energy feature groups for MPR233. (C) PCC values of nonenergy feature groups for MPD276 and MPR233. The last row in each figure shows the performance of integrative feature groups.
Fig 4.
PCC values of different models for LOCOV. (A) Energy feature-based model. (B) Nonenergy feature-based model. (C) Integrative model (PEMPNI).
Fig 5.
Performance of PEMPNI on different mutation subgroups.
(A) PCC values for regression tasks. (B) AUC values for classification tasks. The subgroups that had a small number of samples were not included in this analysis.
Fig 6.
Comparison of PEMPNI and existing methods.
(A) PEMPNI versus other methods for regression tasks. The performance in parentheses was generated from our algorithm trained by the datasets of PremPDI. (B) PEMPNI versus other methods for classification tasks.
Fig 7.
Performance of PEMPNI and other methods on representative protein-DNA complexes.