RankVar: machine learning-based variant ranking and reinterpretation for rare genetic diseases.

Random forest classifier prioritising causative variants from clinical notes and WGS/WES profiles, trained on ~1 million variants

Integrating biological knowledge and phenotype information with variant data remains challenging for identifying disease genes. The authors develop RankVar, a machine learning algorithm prioritising causative variants from clinical notes and WGS/WES profiles. The random forest classifier is trained on ~1 million variants (1000 Genomes + spiked-in pathogenic variants). Tested on several independent datasets, it achieves top-10 accuracy of 90.0% (CHOP), 81.5% (BDB), 46.1% (SSC) and 76.3% (SPARK), with improved performance over existing approaches.

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

A prioritisation tool directly useful for rare disease diagnosis, with good performance on real Mendelian cohorts (top-10 of 90% at CHOP). The drop on autism cohorts (SSC 46%) highlights the limits on complex architectures; it remains to be integrated into pipelines and benchmarked against prioritisers already in routine use.

Analysis by Dr Thibaut Benquey