Fig. 3: Logistic principal component analysis (PCA) and receiver-operating characteristic curves (ROC) allow for variant function prediction. | Genetics in Medicine

Fig. 3: Logistic principal component analysis (PCA) and receiver-operating characteristic curves (ROC) allow for variant function prediction.

From: Computational analysis of 10,860 phenotypic annotations in individuals with SCN2A-related disorders

Fig. 3

(a) ROC performance measurements (area under the curve [AUC]) for phenotypic subgroup comparisons between developmental and epileptic encephalopathy (DEE), benign familial neonatal–infantile seizures (BFNIS), and autism. A darker shade of blue indicates a higher performance for separating between phenotypic groups. (b) The second major principal component (PC2) separates individuals with known loss-of-function (LoF) (blue) and gain-of-function (GoF) (red) variants. (c) Density plot of PC2 across all individuals with known LoF (blue), GoF (red), and unmeasured variants (gray). (d) ROC for PC2 (yellow) shows higher performance for separating GoF from LoF variants. (e) Positive predictive values (PPV) for GoF and LoF variants with PC2 values for individuals with specific variants are highlighted on the graph. Some variants appear twice as phenotypes in individuals with recurrent variants may differ.

Back to article page