Artificial intelligence in molecular diagnostics of genetic diseases - is it smart enough to outsmart humans?

Recent advancements in next generation genome sequencing technologies enable clinical labs and research scientists to generate enormous amount of genomic data that can be further processed for identifying potential causative variants. However, identification of disease-causing variants remains an ongoing challenge, as thousands of variants are present in a single sample, and manual analysis of each case is time-consuming. In recent years, there has been an increase in the number of patients undergoing broad genomic sequencing. However, the number of certified geneticists, bioinformaticians and variant analysts available to interpret sequencing data is insufficient to meet growing expectations. Despite the fact that re-analysis can increase the molecular diagnostic yield by 10-15% for patients going through diagnostic odyssey, in practice it is difficult to perform periodic manual re-analysis of unsolved cases. Therefore, technological solutions are essential in order to address variant interpretation bottleneck. Artificial intelligence-based models take into account various features in order to rank each variant in exome sequencing data of patients with genomic disorders, including the effect of the variant on the protein, allele frequency, segregation data, phenotypic match, publications and databases.

In this talk successes and failures of an AI-based automatic genomic interpretation platform in recognition of disease-causing variants in patients with genetic diseases will be summarized. Situations where clinical input is crucial for genomic variant interpretation will be discussed.