Table 2. Random forest variable importance scores and ‘Out of Bag' (OOB) error rates.
| Variable |
ASD/ID (N=1590) |
DBE (N=1590) |
De Novo (N=1369) |
||||||
|---|---|---|---|---|---|---|---|---|---|
| All | Del | Dup | All | Del | Dup | All | ASD/ID | DBE | |
| ADOS severity score | 0.05 | 0.06 | 0.08 | −0.16 | −0.25 | −0.27 | 0.00 | 0.13 | 0.15 |
| Family type | 0.04 | 0.03 | 0.02 | −0.37 | −0.11 | 0.00 | 0.23 | 0.15 | 0.46 |
| First phrases | 1.15 | 0.32 | 0.32 | −0.42 | 0.61 | −0.31 | 0.67 | 0.51 | 0.55 |
| First words | 0.20 | 0.45 | −0.04 | −0.71 | 0.57 | −0.74 | 0.00 | 0.42 | 0.67 |
| Full-scale IQ | 0.90 | 0.37 | 0.19 | −0.93 | 1.22 | 0.32 | 1.38 | 1.10 | 1.57 |
| Gait disturbance | 0.06 | 0.05 | 0.00 | −0.41 | −0.07 | −0.32 | −0.08 | −0.01 | 0.04 |
| Language delay | 0.94 | 0.54 | 0.05 | −0.09 | −0.13 | -0.06 | 0.22 | 0.29 | 0.39 |
| Maternal age | 0.83 | 0.32 | 0.48 | −0.47 | 1.29 | 0.39 | 0.91 | 0.43 | 0.33 |
| Paternal age | 0.99 | 0.42 | 0.43 | 0.34 | 2.69 | 0.58 | 1.27 | 0.31 | 0.41 |
| Performance IQ | 0.75 | 0.30 | 0.29 | 0.65 | 0.37 | 0.29 | 1.01 | 0.08 | 0.71 |
| Seizures | −0.12 | −0.05 | −0.01 | −0.29 | −0.04 | −0.18 | 0.07 | −0.23 | −0.05 |
| Sex | −0.03 | −0.01 | 0.00 | 0.02 | −0.06 | −0.11 | −0.01 | −0.01 | 0.08 |
| VABS communication | 0.80 | 0.23 | 0.48 | −0.21 | 1.14 | 0.99 | 3.72 | 1.03 | 3.15 |
| VABS composite | 0.59 | 0.62 | 0.53 | 0.29 | 1.14 | 0.52 | 6.29 | 1.79 | 6.34 |
| VABS daily living skills | 0.95 | 0.61 | 0.56 | 0.04 | 0.44 | 0.35 | 4.30 | 1.51 | 3.22 |
| VABS socialization | 0.82 | 0.34 | 0.35 | −0.04 | −0.51 | 1.54 | 3.37 | 0.84 | 2.70 |
| Verbal status | 0.77 | 0.22 | 0.11 | −0.21 | 0.21 | 0.36 | 0.28 | 0.14 | 0.10 |
| Verbal IQ | 2.01 | 0.87 | 0.36 | −0.23 | 1.89 | −0.20 | 1.46 | 0.40 | 1.40 |
| OOB error | 0.07 | 0.03 | 0.04 | 0.52 | 0.27 | 0.32 | 0.08 | 0.02 | 0.05 |
Abbreviations: ADOS, Autism Diagnostic Observation Schedule; ASD/ID, Autism Spectrum Disorder or Intellectual Disability; DBE, differentially brain expressed; Del, Deletion; Dup, Duplication; IQ, intelligence quotient; VABS, Vineland Adaptive Behavior Scales. The top three classifiers in each CNV type are in bold typeface. The OOB error rate reflects classification accuracy over 10 000 permutations. The variable importance score reflects the difference between the OOB error rates when the CNV status is and is not permuted over 10 000 iterations. If the error rates do not increase when the CNV status is permuted, then that variable is a poor classifier. Negative variable importance indicates that the variable hinders classification. Ultimately, variable importance demonstrates which variables have a major role in discriminating participants with and without CNVs.