Table 2.
Performance of four models used in developing NGSS1 and NGSS2
| Number of genes selected | Naïve RSS | Naïve correlation | Naïve misclassified subjects (out of 106) |
CV RSS | CV Correlation | CV prediction accuracy | CV misclassified subjects (out of 106) |
|
|---|---|---|---|---|---|---|---|---|
| Model 1 | 39 genes | 1.234 | 0.909 | 15 | 2.743 | 0.797 | 77.4% | 24 |
| Model 2* | 41genes | 0.884 | 0.935 | 9 | 2.681 | 0.813 | 89.6% | 11 |
| Model 3 | 42 genes | 0.920 | 0.933 | 13 | 2.119 | 0.844 | 78.3% | 23 |
| Model 4** | 13 genes | 2.549 | 0.800 | 16 | 3.215 | 0.741 | 84.0% | 17 |
Naïve and CV RSS are the mean residual sums of squares of the predictive model in the original and cross-validation analyses, respectively. Correlation are the Pearson correlation coefficient between the predicted severity scores and the clinically defined GRSS. Prediction accuracy is the percentage of correctly predicted mild (NGSS ≤ 3.5) or severe (NGSS > 3.5) symptoms, compared with the same phenotype defined by the GRSS (mild: GRSS ≤ 3.5; severe: GRSS > 3.5)
*Designated NGSS1. **Designated NGSS2