Machine learning-based model for prediction of carbamazepine- and allopurinol-induced severe cutaneous adverse reactions in Vietnamese.

Machine learning models on exome (WES) data to predict SCAR risk, beyond the limited PPV of HLA-B15:02 and HLA-B**58:01 alone

The low positive predictive value of HLA-B15:02 and HLA-B**58:01 for stratifying carbamazepine- and allopurinol-induced severe cutaneous adverse reactions (SCARs) calls for better models. The authors genotype 249 patients and controls by WES (75 cases / 73 controls for allopurinol, 48 / 53 for carbamazepine) and apply eight machine learning models. For allopurinol, Random Forest and Extra Tree models reach a mean accuracy of 99.67%; for carbamazepine, Linear SVC achieves a mean AUC of 86%.

Synthesis written by Geno'X. For the full original abstract, please refer to the source publication.

An interesting approach to move beyond the limited PPV of individual HLA-B alleles in predicting SCARs. The reported performance (99.67%) should be treated cautiously given the very small samples and overfitting risk: it is a promising proof of concept requiring external validation before any clinical use.

Analysis by Dr Thibaut Benquey