AI and Rare Diseases ── Platforms ── 2026-11-16
Rare-disease diagnosis and drug discovery platforms and AI
- Abstract
BACKGROUND: Large language models (LLMs) are increasingly used in medical education, but their ability to analyze pedigree-based genetics problems is not well defined. This exploratory study aimed to compare the accuracy of four consumer LLMs on image-based pedigree questions and to describe the sources of their errors. METHODS: Nine pedigree cases comprising 19 sub-questions were submitted as images on May 18, 2026, to the free consumer versions of ChatGPT (GPT-5.5), Gemini (Gemini 3 Flash), Claude (Claude Sonnet 4.6), and the Microsoft Copilot app, once each in a new chat, without retries or follow-up prompts. The primary measure was accuracy, defined as the number and percentage of sub-questions answered correctly against a predefined answer key reviewed by a medical geneticist. Item-level scores were re-checked during revision using uniform rules. Incorrect answers were retrospectively classified as perception/structure, reasoning/knowledge, mixed or indeterminate errors, or as reported image-access failures. RESULTS: ChatGPT and Claude each answered 14 of 19 sub-questions correctly (73.7%), Copilot 11 (57.9%), and Gemini three (15.8%). Twelve of Gemini's 16 incorrect answers followed its report that it could not see or load the image. Claude's errors in the double-cousin pedigrees reflected misreading of the mating lines, whereas ChatGPT's mainly involved assigning the degree of relationship. In the single atypical pedigree (dominantly inherited hemoglobin S-Oman), no model identified the variant; Claude described dominant features but attributed them to pseudodominance because sickle cell disease is recessive. CONCLUSIONS: In this small, single-attempt task set without statistical comparison, consumer LLMs showed variable accuracy, and errors arose both from misreading pedigrees and from misapplying genetic principles. These results do not establish a ranking of the models or an educational benefit. Consumer AI assistants should complement, not replace, formal instruction and expert supervision in medical genetics.
Journal IF-equivalent: 1.9 (OpenAlex 2-year mean citedness, value as of 2026-11-15, retrieved 2026-11-16; not the official Clarivate IF)Reference: Abduljabbar SS, Alqarni M, Habara A. Consumer Artificial Intelligence for Pedigree Analysis: A Comparative Study in Medical Genetics. Cureus. 2026 Oct 15 [Epub ahead of print]. doi:10.7759/cureus.117953. PMID: 43034519.Checked: Abstract only