Table 1.
Improvements in prediction accuracy.
| AA-type | Property | GAA-type | NN Prediction | Improvement |
| arg | I | 9.7% | 13.6% | 40.2% |
| asn | I | 8.3% | 13.3% | 60.2% |
| asp | I | 8.2% | 13.7% | 67.1% |
| gln | I | 8.7% | 13.3% | 52.9% |
| glu | I | 10.6% | 15.1% | 42.5% |
| his | I | 7.7% | 12.0% | 55.8% |
| lys | I | 9.7% | 14.9% | 53.6% |
| ser | I | 10.9% | 16.5% | 51.4% |
| thr | I | 9.1% | 15.3% | 68.1% |
| gly | - | 15.0% | 16.2% | 8.0% |
| ala | O | 12.4% | 17.3% | 39.5% |
| cys | O | 10.5% | 14.0% | 33.3% |
| ile | O | 14.3% | 15.3% | 7.0% |
| leu | O | 12.5% | 16.0% | 28.0% |
| met | O | 9.8% | 12.3% | 25.5% |
| phe | O | 10.4% | 12.6% | 21.2% |
| pro | O | 21.4% | 27.0% | 26.2% |
| trp | O | 13.5% | 15.2% | 12.6% |
| tyr | O | 9.4% | 12.0% | 27.7% |
| val | O | 13.7% | 14.2% | 3.6% |
Prediction accuracy of the neural network is compared to a lower bound derived from a purely statistical analysis of the data set. 'O' and 'I' in the "property" column denotes hydrophobic and hydrophilic residues respectively.