Method | Exposure | Train | CB513/CB511 | Method |
|---|
Ahmad | ASA | - | 0.48 | ANN |
Yuan | ASA | - | 0.52 | SVR |
Nguyan | ASA | - | 0.66 | Two-Stage SVR |
Real-SPINE | ASA | 0.74 | 0.73 | ANN |
Real-SPINE | RSA | - | 0.70 | ANN |
NetSurfP | ASA | 0.75 | 0.72 | ANN |
NetSurfP | RSA | 0.72 | 0.70 | ANN |
- Performances are shown for 5 different approaches to predict absolute and relative (RSA) surface accessibility. Methods included in the benchmark are Ahmad: [5], Yuan: [20], Nguyen: [24], Real-SPINE: [22], NetSurfP: This work. Train gives the training performance, and CB513/CB511 gives the evaluation performance on the CB513 data set. Train performance of the Real-SPINE method and evaluation performances for the Ahmad, Yuan, and Nguyen method are taken from the corresponding publications. ANN = Artificial neural networks, SVR = Support vector regression. Pearson's correlation coefficients (PCC) are shown for all methods based on the absolute surface exposure of an amino acid. Also, PCC values are given for relative surface exposure for the two methods NetSurfP and Real-SPINE.
