Phenotype-first patient matching with SimPheny identifies diagnostic candidates beyond curated gene associations

Isabelle Cooperstein, Alistair Ward, Paul Avillach, Emerson Lebleu, Barry Moore, Rebecca C Spillmann, Vandana Shashi, Undiagnosed Diseases Network, Thomas GabormedRxiv 2026 · DOI

TL;DR

SimPheny is a novel gene prioritization algorithm that identifies rare disease diagnoses by matching undiagnosed patients to a reference cohort of real diagnosed cases rather than static "textbook" disease models. By prioritizing phenotypic similarity between patients first, it successfully identifies causative genes even when those gene-disease associations are missing or incomplete in curated databases like OMIM.

What Problem Does This Paper Address?

Rare monogenic diseases affect over 400 million people worldwide, yet identifying the specific genetic cause for a patient remains a "diagnostic odyssey" that can last years. Current computational tools typically rely on curated databases (like OMIM or Orphanet) that link specific genes to specific symptoms (phenotypes).

However, these tools face a major "cold start" problem:

  1. Incomplete Knowledge: Many rare disease genes haven't been discovered or documented yet.
  2. Atypical Presentations: Real patients rarely look exactly like the "idealized" descriptions in textbooks.
  3. Static Models: Curated databases are slow to update, meaning newly discovered gene-disease relationships are often missed by standard pipelines.

Methods

The authors introduce SimPheny, which shifts the focus from "Patient-to-Database" matching to "Patient-to-Patient" matching.

  • PhenoSimJaccard: To compare symptoms, the authors developed a modified Jaccard index. Unlike a standard overlap count, this metric uses the Human Phenotype Ontology (HPO) to reward matches on highly specific symptoms and shared ancestral terms.
  • Statistical Scoring: SimPheny calculates a score based on two empirical p-values:
  • pheno p: The probability that the phenotypic similarity between two patients happened by chance.
  • gene p: The probability that a specific gene would appear in a patient's candidate list by chance (adjusting for "frequent flyer" genes that are large or highly variable).
  • Empirical Brown’s Method (EBM): Since phenotypic similarity and gene presence are often correlated, the tool uses EBM to combine these p-values into a single, robust "SimPheny Score."
  • Workflow: The tool takes a list of a patient's symptoms and a list of their candidate gene variants (filtered by tools like Exomiser) and searches a reference database for diagnosed patients with similar profiles.

Datasets

The study primarily utilized data from the Undiagnosed Diseases Network (UDN), a high-complexity cohort of 2,214 participants.

  • Benchmarking Set: 404 diagnosed probands (used to test if the tool could "re-diagnose" them).
  • Discovery Set: 1,447 undiagnosed probands.
  • Expanded Reference Cohorts: To test scalability, the authors integrated external data from the Phenopacket Store, ClinVar, and DECIPHER, expanding the reference pool to over 23,000 individuals.

Key Insights and Ideas

  • Real World > Idealized Models: The authors found that real patients with the same diagnosis are often more phenotypically similar to each other than they are to the "gold-standard" curated models in databases.
  • Bypassing the Curator: Because SimPheny matches patient-to-patient, it can identify a diagnosis for a "novel" disease as soon as one other person in the network is diagnosed, even if the discovery hasn't been published or added to OMIM yet.
  • Scaling without Noise: A common fear in bioinformatics is that adding more (and potentially "messier") data will dilute results. SimPheny proved the opposite: expanding the reference cohort significantly increased the number of genes the tool could "see" (diagnostic reach) without sacrificing the precision of high-confidence matches.

Key Results

  • Superior Ranking: In benchmarks, SimPheny ranked the true diagnostic gene in the top five candidates for 96.2% of cases, outperforming state-of-the-art tools like AI-MARRVEL (84.0%).
  • Solving the "Unsolvable": For genes with zero documented HPO associations in curated databases, SimPheny still ranked the correct gene as the #1 candidate.
  • Clinical Success: When applied to 1,447 unsolved UDN cases, SimPheny generated 18 high-confidence leads. Clinical review confirmed that nearly half (44.4%) were definitive diagnoses, including a "cold case" involving the gene RPL13 that had remained unsolved for five years.

Limitations and Future Work

  • Reference Dependency: SimPheny is a "matching" tool; it cannot identify a gene if no one else in the reference database has been diagnosed with it yet.
  • Upstream Reliance: The tool requires a candidate gene list from a variant-caller. If the causative variant is missed during initial sequencing or filtering (e.g., a complex structural variant), SimPheny cannot recover it.
  • Future Directions: The authors are working on "undiagnosed-to-undiagnosed" matching, which would allow researchers to find clusters of similar patients to define entirely new genetic syndromes in real-time.

Summary updated

Abstract

Diagnostic tools for rare diseases typically rely on curated gene-phenotype associations and static disease models, limiting their effectiveness in cases with atypical presentations or previously uncharacterized disorders. To address these limitations, we present SimPheny, a phenotype-first algorithm for gene prioritization that operates independently of documented gene-phenotype associations. SimPheny identifies phenotypically similar diagnosed patients by comparing an undiagnosed patient's disease presentation to a reference cohort of diagnosed cases, and returns gene hypotheses by matching the undiagnosed patient's candidate gene list to the causative genes of similar patients using a statistical scoring model. Evaluated in diagnosed probands from the Undiagnosed Diseases Network (UDN) with the true diagnostic gene blinded, SimPheny consistently ranked the diagnostic gene among the top five candidates, outperforming existing tools, particularly for genes with limited gene-phenotype association data. When applied to previously unsolved UDN cases, clinical review confirmed that SimPheny's high-confidence causative gene predictions were diagnostic in nearly half of the analyzed cases. As the size of the diagnosed reference cohort increases, SimPheny's diagnostic reach expands without sacrificing ranking performance. By leveraging real patient data rather than curated models, SimPheny provides a generalizable, scalable framework for improving diagnostic yield in rare disease cohorts.
BibTeX
@misc{cooperstein2026phenotype,
  author = {Isabelle Cooperstein and Alistair Ward and Paul Avillach and Emerson Lebleu and Barry Moore and Rebecca C Spillmann and Vandana Shashi and Undiagnosed Diseases Network and Thomas Gabor},
  title = {{Phenotype-first patient matching with SimPheny identifies diagnostic candidates beyond curated gene associations}},
  year = {2026},
  doi = {10.64898/2026.01.15.26344236},
  howpublished = {medRxiv},
}
Figure