Table 2. The performance of PLMGraph-Inter when using different sequence identity and fold similarity thresholds to further remove potential redundancies in HomoPDB and HeteroPDB.
| HomoPDB | HeteroPDB | ||||
|---|---|---|---|---|---|
| Count |
Precision
(Top 50 [%]) |
Count |
Precision
(Top 50 [%]) |
||
| Sequence identity (MMSeqs2) |
Original | 400 | 67.3 (60.9) | 200 | 41.4 (37.8) |
| 40% | 341 | 68.7 (62.5) | 160 | 38.6 (35.6) | |
| 30% | 257 | 64.7 (58.7) | 144 | 38.1 (35.1) | |
| 20% | 211 | 63.2 (56.3) | 138 | 38.5 (35.3) | |
| 10% | 211 | 63.2 (56.3) | 138 | 38.5 (35.3) | |
| Fold similarity (TM-align) |
0.9 | 370 | 65.2 (58.3) | 185 | 39.7 (35.7) |
| 0.8 | 281 | 61.8 (53.4) | 153 | 38.1 (34.1) | |
| 0.7 | 179 | 56.5 (45.8) | 126 | 38.8 (34.6) | |
| 0.6 | 124 | 50.4 (39.9) | 102 | 37.4 (34.1) | |
| 0.5 | 70 | 49.6 (41.3) | 83 | 36.5 (34.5) | |
The results using experimental structures are shown outside the parentheses, and the results using the AlphaFold2 predicted structures are shown inside the parentheses.