Table 2 Machine Learning Model Results
Model Input | Model Type | Architecture Diagram | Task(s) | T2D Classification AUC (95% CI) | Estimated PC MSE | Estimated PC R2 | T2D Classification from Estimated PCs AUC (95% CI) | T2D Classification from Alternate Input |
|---|---|---|---|---|---|---|---|---|
PCs | NN | Fig. 1a | T2D | 0.56 (0.54–0.56) | --- | --- | --- | --- |
Genotype | NN | Fig. 1b | PCs, T2D from PCs (stop gradient) | --- | 0.27 | 0.73 | 0.55 (0.54–0.56) | --- |
PC adjusted PRS | LR | --- | T2D | 0.57 (0.56–0.58) | --- | --- | --- | --- |
PRS-CS | LR | --- | T2D | 0.59 (0.58–0.60) | --- | --- | --- | --- |
LDpred2 | LR | --- | T2D | 0.63 (0.62–0.64) | --- | --- | --- | --- |
Genotype | NN | Fig. 1c | T2D, PCs, T2D from PCs | 0.66 (0.65–0.67) | 0.38 | 0.62 | 0.56 (0.55–0.57) | --- |
Genotype | NN | Fig. 1d | T2D, PCs (stop gradient) | 0.65 (0.64–0.66) | 1.70 | <0 | --- | --- |
Genotype | NN | Fig. 1e | T2D, PCs (adversarial) | 0.66 (0.65–0.67) | 2.32 | <0 | --- | --- |
CID Alternate: Genotype | CNN | Fig. 1f | T2D, T2D from genotype, PCs (stop gradient) | 0.62 (0.61–0.63) | 0.69 | <0 | --- | 0.63 (0.62 – 0.64) |
CID Alternate: Genotype | CNN | Fig. 1g | T2D from CID, T2D from genotype (stop gradient), PCs (stop gradient) | 0.65 (0.64–0.66) | 0.66 | <0 | --- | 0.54 (0.53 – 0.55) |
CID Alternate: Genotype | CNN | Fig. 1h | T2D from CID, T2D from genotype (adversarial), PCs (adversarial) | 0.59 (0.58–0.60) | 3.30 | <0 | --- | 0.50 (0.50 – 0.50) |
Genotype Alternate: CID | CNN | Fig. 1i | T2D from genotype, T2D from CID (adversarial), PCs (adversarial) | 0.57 (0.56–0.58) | 3.37 | <0 | --- | 0.50 (0.49 – 0.51) |
h + i intermediate layer output | NN | --- | T2D | 0.61 (0.60–0.62) | --- | --- | --- | --- |
