PPI-hotspotID for detecting protein–protein interaction hot spots from the free protein structure
- Institute of Biomedical Sciences, Academia Sinica, Taiwan
Figures
Figure 1
Interface vs. noninterface PPI-hot spots.
(Top left) The X-ray structure (PDB: 1IAR) of interleukin-4 (gray) in complex with interleukin-4 receptor subunit alpha (wheat) with five PPI-hot spots; interface PPI-hot spots (E13, R92, and N93) …
Figure 2 with 1 supplement
Evaluation of the predicted CPEB2-interacting amino acid residues in eEF2.
(a) Salient features of mouse eEF2, showing the various domains and the mutated amino acids in domain V. mut 1, G778A, T779A, and P780A; mut 2, F794A; mut 3, D815A. (b) Reciprocal …
-
Figure 2—source data 1
Containing uncropped images of the membranes for Figure 2b, c and e and phosphoimager file for Figure 2d.
- https://cdn.elifesciences.org/articles/96643/elife-96643-fig2-data1-v1.docx
-
Figure 2—source data 2
Containing the original files of the full raw unedited blots.
- https://cdn.elifesciences.org/articles/96643/elife-96643-fig2-data2-v1.zip
-
Figure 2—source data 3
Listing primer sequences.
- https://cdn.elifesciences.org/articles/96643/elife-96643-fig2-data3-v1.docx
Figure 2—figure supplement 1
Uncropped immunoblot images.
The uncropped images for Figure 2b, c, and e are shown with indicated molecular weight marker.
Figure 3
Interface and noninterface PPI-hot spots of Golgi resident protein (GCP60).
(Top) The structure (PDB 2N73) (Wright et al., 2024) of GCP60 (green) in complex with PI4K-β (cyan) with the GCP60–PI4K-β interface encircled. (Bottom) The four experimentally known PPI-hot spots of …
Figure 4
Based on the free CRADD X-ray structure (PDB 2O71-A), PPI-hotspotID predicted G128 as a PPI-hot spot for CRADD–CRADD interactions.
(Left) The structure (PDB 2OF5) (Park et al., 2007) of seven CRADD proteins in complex with fuve PIDD proteins. The circle shows the CRADD–CRADD interface between chains C (cyan) and G (orange), …
Tables
Table 1
Performance of the PPI-hotspotID vs. FTMap and SPOTONE.
| Method | PPI-hotspotID | FTMap | SPOTONE | |
|---|---|---|---|---|
| TP | 278 | 30 | 40 |  |
| FN | 136 | 384 | 374 | |
| TN | 417 | 487 | 481 | |
| FP | 87 | 17 | 23 | |
| Sensitivity | 0.67 | 0.07 | 0.10 | |
| Precision | 0.76 | 0.64 | 0.64 | |
| F1 | 0.71 (0.66)* | 0.13 | 0.17 | |
| Specificity | 0.83 | 0.97 | 0.95 |
-
Each method was tested using the same dataset comprising 414 experimentally known PPI-hot spots (TP + FN) and 504 PPI-nonhot spots (TN + FP).
-
TP = true positive; FP = false positive; TN = true negative; FN = false negative.
-
*
The F1 score in parentheses corresponds to the validation F1 score.
Table 2
Performance of AlphaFold-Multimer, PPI-hotspotID, and their combination for 48 ‘unsolved’ complex structures.
| Method | AlphaFold2-Multimer | PPI-hotspotID | AlphaFold2-Multimer+PPI-hotspotID |
|---|---|---|---|
| TP | 37 | 52 | 63 |
| FN | 53 | 38 | 27 |
| TN | 35 | 29 | 24 |
| FP | 10 | 16 | 21 |
| Sensitivity | 0.41 | 0.58 | 0.70 |
| Precision | 0.79 | 0.77 | 0.75 |
| F1 | 0.54 | 0.66* | 0.72 |
| Specificity | 0.78 | 0.64 | 0.53 |
-
Each method was tested using the same dataset comprising 90 experimentally known PPI-hot spots (TP+FN) and 45 PPI-nonhot spots (TN+FP) in 48 protein complexes with no known structures.
-
TP = true positive; FP = false positive; TN = true negative; FN = false negative.
-
*
No validation F1 score is provided since AutoGluon was used to train an ensemble of machine-learning models on a dataset that excludes the 48 ‘unsolved’ complex structures (see text).
Additional files
-
MDAR checklist
- https://cdn.elifesciences.org/articles/96643/elife-96643-mdarchecklist1-v1.pdf
-
Source data 1
Dataset of experimentally confirmed PPI-hot spots and PPI-nonhot spots with free protein structures.
- https://cdn.elifesciences.org/articles/96643/elife-96643-data1-v1.xlsx
-
Source data 2
Dataset of experimentally confirmed PPI-hot spots with both free and bound protein structures.
- https://cdn.elifesciences.org/articles/96643/elife-96643-data2-v1.xlsx
