Table 1.
Comparison of the performance for the different feature extraction methods. * AUC confidence Interval calculated using a parametric Gaussian approximation for the predictive output. ** A 95% confidence interval has been computed according to the binomial distribution, taking into account the finite number of samples used for the estimation of the proportion.
| Area of the RIC | RIC | Full Matrix | Peak Table | |
|---|---|---|---|---|
| AUC in Cross-Validation | 0.79 | 0.99 | 0.96 | 0.97 |
| CR in Cross-Validation | 77% | 98% | 95% | 95% |
| AUC in External Validation | 0.81 | 0.97 | 0.98 | 0.96 |
| 95% Conf. Interval for AUC * | 0.65–0.91 | 0.88–0.99 | 0.89–0.99 | 0.88–0.99 |
| CR in External Validation | 75% | 94% | 93% | 94% |
| 95% Conf. Interval for CR ** | 61–86% | 83–99% | 83–98% | 84–99% |
| Computational Cost | Very Low | Very Low | High | Moderate |
| Implementation Difficulty | Easy | Easy | Moderate | Difficult |
| Chemical Information | None | Moderate | Very High | High |
| Number of Features | 1 | 3500 | 700,000 | 189 |