What question did this study set out to answer?

The study aims to improve the classification of pathogenic variants using an AI tool called MIMI Ranker based on extensive genetic data.

March 8, 2026Open Access

P551: MIMI Ranker: An AI-powered tool for pathogenic variant prioritization tested on a diverse dataset of 16K clinical exomes and genomes*

ABAmber BegtrupGenVec HGHakon Gudbjartsson MSMaria Guillen Sacoto

Key Points

The study aims to improve the classification of pathogenic variants using an AI tool called MIMI Ranker based on extensive genetic data.
Evaluated the performance of the MIMI Ranker AI tool on a dataset of 16,000 clinical exomes and genomes.
Applied Bayesian analysis to assess the reclassification of variants over a two-year period.
Monitored the impact of integrating literature-derived evidence on variant classifications.
Reclassification rate of 39.2% for clinically actionable categories among variants after two years.
Out of 390 VUS, 175 (44.9%) were reclassified, including upgrades to pathogenic (P) and likely pathogenic (LP).
New functional evidence from literature was critical, impacting 169 variants and facilitating accurate and updated classifications.

Abstract

Overall, 463 (36.9%) variants were reclassified, while 791 (63.1%) retained their classification.The largest clinically relevant effect was observed within the VUS group.Namely, 175 of 390 VUS (44.9%) were reclassified: 36 (9.2%) VUS were upgraded to P, 117 (30%) VUS were upgraded to LP, 19 (4.9%) VUS were downgraded to LB/B, and 3 (0.8%) VUS to conflicting.Among LP variants, 249 of 488 (51.0%) were upgraded to P. Among 12 LB variants, 2 were downgraded to B, 1 to conflicting, and 5 reverted to VUS.Bayesian analysis of the reclassified VUS (n=175) revealed a clear relationship between VUS subcategories and reclassification: variants classified as 'Low' (n=24) largely shifted to B/LB (n=17) while 7 variants shifted to LP; 'Mid' variants (n=80) exhibited an upward shift with 76 variants shifting to P/LP, 2 variants to LB/B, and 2 conflicting; 'High' variants (n=71) exclusively shifted to P/LP (n=70) with only 1 variant classified as 'conflicting'.Bayesian analysis of VUS that were not reclassified in two years (n=215) showed an upwards trend towards a higher VUS subcategory.The greatest impact on reclassification came from literature-derived data, with 169 variants gaining new functional evidence and 86 new clinical evidence, underscoring the value of literature evidence.Conclusion: Over a two-year period, 39.2% of GLA VUS were reclassified into clinically actionable categories, primarily driven by new functional and clinical evidence from the literature.Bayesian stratification effectively predicted future pathogenic classifications, enabling prioritization of variants for reevaluation.These results underscore the importance of continuously integrating new information into variant classification workstreams and ensuring clinician awareness of updated interpretations to support accurate diagnosis and personalized management.Integrating variant re-evaluation into clinical workflows will be essential for delivering dynamic, evidence-driven patient care in Fabry disease.

Ask AI

Helpful

Bookmark

View Full Paper