Assessment of In-Frame Indel Variants in an Unsolved Cohort of Inherited Retinal Diseases Using Machine Learning.

The standard for in silico pathogenicity prediction of in-frame insertions and deletions (indels) is less established compared to other types of variations. We aimed to systematically assess the performance of in silico machine learning (ML) tools on a patient cohort with inherited retinal diseases (IRDs). The performance of four ML tools (CADD, FATHMM-indel, VEST4, and MetaRNN-indel) was compared. Among them, MetaRNN-indel showed the best overall results. MetaRNN-indel was then applied to 1013 unsolved IRD patients, identifying two likely pathogenic causal variants in two unrelated IRD patients by confirming clinical phenotypes. Hence, our findings indicate that reliable prediction of the pathogenicity of in-frame indels can be achieved using existing ML tools with proper evaluation and tuning.