Pediatrics · Medical Literature
LLM-Assisted Reanalysis of Unsolved Rare Disease Genomes Increases Diagnostic Yield.
Abstract
Rare and undiagnosed genetic disorders affect millions of patients globally, and many patients endure years of inconclusive testing. Conventional genomic interpretation can be insufficiently sensitive and costly and is rarely repeated as knowledge evolves. We conducted a retrospective multicohort reanalysis using a large language model (LLM)-assisted workflow that ingests clinician notes, Human Phenotype Ontology (HPO) terms, and a filtered variant table to propose explanation-rich candidate hypotheses for expert adjudication under American College of Medical Genetics and Genomics and Association for Molecular Pathology criteria. A diagnosis was defined a priori as a variant classified as pathogenic or likely pathogenic, confirmed in a Clinical Laboratory Improvement Amendments-certified laboratory, and returned to families. Secondary outputs included "rediscoveries" of externally established diagnoses not yet available locally and hypothesis generation signals. Across four cohorts, new local diagnoses were made in 10 of 100 rare disease neurodevelopmental cases (10.0%, [exact binomial: 95% confidence interval (CI), 4.9 to 17.6]), 4 of 61 neuromuscular cases (6.6%, [CI, 1.8 to 16.0]), 2 of 200 cases of sudden unexpected death in pediatrics (1.0% [CI, 0.1 to 3.6]), and 2 of 15 early psychosis cases (13.3% [CI, 1.7 to 40.5]) for an overall diagnostic yield of 18 of 376 (4.8%, [CI, 2.9 to 7.5]). We identified seven rediscoveries in which pathogenic or likely pathogenic findings had been established externally but were not available in the local research record at the time of review. In one case, the model's synthesis of genotype-quality patterns and phenotype concordance triaged a putative 22q11.2 deletion that was subsequently confirmed by whole-genome sequencing. The workflow also generated testable biological hypotheses, including a candidate association between the sphingosine-1-phosphate receptor 1 gene (S1PR1) and vitiligo. In retrospective reanalysis, an explanation-first LLM applied to routine HPO terms and variant tables produced clinically relevant gains in diagnostic yield, surfaced overlooked pathogenic findings, and generated biologically grounded hypotheses. These results motivate prospective multicenter evaluation with predefined end points, calibration reporting, and comparator baselines. (Funded by the U.S. National Institute of Child Health and Human Development and others.).
Clinical relevance
Millions of patients with rare genetic disorders remain undiagnosed after conventional genomic workup, partly because reanalysis as medical knowledge evolves is resource-intensive and rarely performed systematically. This study addresses that gap by deploying an LLM-assisted reanalysis workflow across four distinct pediatric cohorts—neurodevelopmental, neuromuscular, sudden unexpected pediatric death, and early psychosis—representing a broad, clinically heterogeneous rare-disease population. Given the accelerating pace of variant reclassification and gene-disease discovery, clinicians and genomics programs need scalable, knowledge-updating tools to recover diagnoses from previously inconclusive sequencing data.
Key findings
1. Overall diagnostic yield from LLM-assisted reanalysis across all four cohorts was 18 of 376 cases (4.8%; 95% CI, 2.9–7.5%), meeting the pre-specified definition of pathogenic or likely pathogenic variant confirmed in a CLIA-certified laboratory and returned to families. 2. Diagnostic yield by cohort: rare disease neurodevelopmental cases 10/100 (10.0%; 95% CI, 4.9–17.6%); neuromuscular cases 4/61 (6.6%; 95% CI, 1.8–16.0%); sudden unexpected death in pediatrics 2/200 (1.0%; 95% CI, 0.1–3.6%); early psychosis cases 2/15 (13.3%; 95% CI, 1.7–40.5%). 3. Seven 'rediscoveries' were identified in which pathogenic or likely pathogenic findings had been externally established but were not yet available in the local research record, highlighting a distinct category of diagnostic recovery beyond de novo variant identification. 4. In at least one case, the model's integration of genotype-quality patterns with phenotype concordance triaged a putative 22q11.2 deletion that was subsequently confirmed by whole-genome sequencing, demonstrating the workflow's capacity for structural variant hypothesis generation. 5. The workflow generated novel biologically grounded hypotheses, including a candidate gene-disease association between S1PR1 (sphingosine-1-phosphate receptor 1) and vitiligo, illustrating potential utility beyond immediate diagnostic resolution. 6. The LLM ingested clinician notes, HPO terms, and filtered variant tables to produce explanation-rich candidate hypotheses adjudicated under ACMG/AMP criteria, embedding interpretive transparency into the diagnostic pipeline.
Practice implications
Genomics programs managing unsolved rare disease cases—particularly in pediatric neurodevelopmental and early psychosis cohorts where yield was highest (10.0% and 13.3%, respectively)—should consider structured LLM-assisted reanalysis as a scalable adjunct to periodic case review, especially for patients with previously inconclusive exome or genome sequencing. Clinicians should ensure that phenotypic data are captured systematically in HPO terms and that variant tables are maintained in machine-readable format, as these inputs directly enable LLM-assisted workflows and maximize reanalysis sensitivity. The authors appropriately call for prospective multicenter trials with predefined endpoints, calibration reporting, and comparator baselines before widespread clinical adoption; in the interim, this approach is best deployed within research or quality-improvement frameworks with expert genomicist adjudication retained as the final interpretive authority.