New AI Tool Aids Diagnosis of Rare Diseases by Analyzing Symptoms
January 29, 2024Researchers at KAUST have developed an AI tool called STARVar that could improve diagnosis of rare diseases. STARVar analyzes patient symptoms, genomic data, and scientific literature to identify disease-linked genetic variants.
Unlike other tools, STARVar can flexibly interpret symptom descriptions in both standardized terms and natural language. This allows for more nuanced analysis of real-world clinical data.
Lead author Senay Kafkas explains that by leveraging symptoms expressed in free text, STARVar can better prioritize variants associated with a patient’s condition.
STARVar was created by Robert Hoehndorf’s team. It combines genomic information, literature evidence, and crucially, the ability to understand symptoms described in a doctor’s own words.
The researchers highlight STARVar’s adaptability to real-world clinical data as a key advantage over methods requiring standardized vocabularies.
By employing AI to integrate diverse data sources, STARVar represents an important advance in unraveling the genetics of rare and poorly understood diseases. The tool demonstrates the power of AI to comprehend nuanced human language and enhance genomic analysis.
STARVar Outperforms Other Tools in Diagnosing Real Patients
The researchers tested STARVar on genomic datasets from Saudi Arabian and international patients. It consistently ranked the correct disease variant highest among hundreds of candidates.
This demonstrates STARVar’s superior performance over tools relying on rigid symptom vocabularies. The algorithm leverages flexible text analysis to pinpoint culprit mutations.
The team applied STARVar to diagnose a Saudi girl with various symptoms. Out of nearly 800 variants, it accurately identified the pathogenic mutation in the MMP2 gene responsible for her condition.
This real-world case highlights STARVar’s clinical utility in interpreting genomic data in the context of complex symptoms.
STARVar is now freely available online. The researchers hope clinical geneticists will incorporate this novel analytic approach into diagnostic workflows.
By combining literature evidence, genomes, and crucially, nuanced symptom data, STARVar represents a powerful new AI asset for solving rare disease cases.
Kafkas, S., et al. (2023). Starvar: symptom-based tool for automatic ranking of variants using evidence from literature and genomes. BMC Bioinformatics. doi.org/10.1186/s12859-023-05406-w.