Large language models (LLMs) are increasingly used medicine for diverse applications including differential diagnostic support. The training data used to create LLMs such as the Generative Pretrained Transformer (GPT) predominantly consist of English-language texts, but LLMs could be used across the globe to support diagnostics if language barriers could be overcome. Initial pilot studies on the utility of LLMs for differential diagnosis in languages other than English have shown promise, but a large-scale assessment on the relative performance of these models in a variety of European and non-European languages on a comprehensive corpus of challenging rare-disease cases is lacking. We created 4917 clinical vignettes using structured data captured with Human Phenotype Ontology (HPO) terms with the Global Alliance for Genomics and Health (GA4GH) Phenopacket Schema. These clinical vignettes span a total of 360 distinct genetic diseases with 2525 associated phenotypic features. We used translations of the Human Phenotype Ontology together with language-specific templates to generate prompts in English, Chinese, Czech, Dutch, French, German, Italian, Japanese, Spanish, and Turkish. We applied GPT-4o, version gpt-4o-2024-08-06, and the medically fine-tuned Meditron3-70B to the task of delivering a ranked differential diagnosis using a zero-shot prompt. An ontology-based approach with the Mondo disease ontology was used to map synonyms and to map disease subtypes to clinical diagnoses in order to automate evaluation of LLM responses. For English, GPT-4o placed the correct diagnosis at the first rank 19.9% and within the top-3 ranks 27.0% of the time. In comparison, for the nine non-English languages tested here the correct diagnosis was placed at rank 1 between 16.9% and 20.6%, within top-3 between 25.4% and 28.6% of cases. The Meditron3 model placed the correct diagnosis within the first 3 ranks for 20.9% of cases in English and between 19.9% and 24.0% for the other nine languages. The differential diagnostic performance of LLMs across a comprehensive corpus of rare-disease cases was largely consistent across the ten languages tested. This suggests that the utility of LLMs in clinical settings may extend to non-English clinical settings. NHGRI 5U24HG011449, 5RM1HG010860, R01HD103805 and R24OD011883. P.N.R. was supported by a Professorship of the Alexander von Humboldt Foundation; P.L. was supported by a National Grant (PMP21/00063 ONTOPREC-ISCIII, Fondos FEDER). C.M., J.R. and J.H.C. were supported in part by the Director, Office of Science, Office of Basic Energy Sciences, of the US Department of Energy (Contract No. DE-AC0205CH11231).

SETD2 belongs to the family of histone methyltransferase proteins and has been associated with three nosologically distinct entities with different clinical and molecular features: Luscan–Lumish syndrome (LLS), intellectual developmental disorder, autosomal dominant 70 (MRD70), and Rabin–Pappas syndrome (RAPAS). LLS [MIM #616831] is an overgrowth disorder with multisystem involvement including intellectual disability, speech delay, autism spectrum disorder (ASD), macrocephaly, tall stature, and motor delay. RAPAS [MIM #6201551] is a recently reported multisystemic disorder characterized by severely impaired global and intellectual development, hypotonia, feeding difficulties with failure to thrive, microcephaly, and dysmorphic facial features. Other neurologic findings may include seizures, hearing loss, ophthalmologic defects, and brain imaging abnormalities. There is variable involvement of other organ systems, including skeletal, genitourinary, cardiac, and potentially endocrine. Three patients who carried the missense variant p.Arg1740Gln in SETD2 were reported with a moderately impaired intellectual disability, speech difficulties, and behavioral abnormalities. More variable findings included hypotonia and dysmorphic features. Due to the differences with the two previous phenotypes, this association was then named intellectual developmental disorder, autosomal dominant 70 [MIM 620157]. These three disorders seem to be allelic and are caused either by loss-of-function, gain-of-function, or missense variants in the SETD2 gene. Here we describe 18 new patients with variants in SETD2, most of them with the LLS phenotype, and reviewed 33 additional patients with variants in SETD2 that have been previously reported in the scientific literature. This article offers an expansion of the number of reported individuals with LLS and highlights the clinical features and the similarities and differences among the three phenotypes associated with SETD2.

Snijders Blok–Campeau syndrome (SNIBCPS, OMIM# 618205) is an extremely infrequent disease with only approximately 60 cases reported so far. SNIBCPS belongs to the group of neurodevelopmental disorders (NDDs). Clinical features of patients with SNIBCPS include global developmental delay, intellectual disability, speech and language difficulties and behavioral disorders like autism spectrum disorder. In addition, patients with SNIBCPS exhibit typical dysmorphic features including macrocephaly, hypertelorism, sparse eyebrows, broad forehead, prominent nose and pointed chin. The severity of the neurological effects as well as the presence of other features is variable among subjects. SNIBCPS is caused likely by pathogenic and pathogenic variants in CHD3 (Chromodomain Helicase DNA Binding Protein 3), which seems to be involved in chromatin remodeling by deacetylating histones. Here, we report 20 additional patients with clinical features compatible with SNIBCPS from 17 unrelated families with confirmed likely pathogenic/pathogenic variants in CHD3. Patients were analyzed by whole exome sequencing and segregation studies were performed by Sanger sequencing. Patients in this study showed different pathogenic variants affecting several functional domains of the protein. Additionally, none of the variants described here were reported in control population databases, and most computational predictors suggest that they are deleterious. The most common clinical features of the whole cohort of patients are global developmental delay (98%) and speech disorder/delay (92%). Other frequent features (51–74%) include intellectual disability, hypotonia, hypertelorism, abnormality of vision, macrocephaly and prominent forehead, among others. This study expands the number of individuals with confirmed SNIBCPS due to pathogenic or likely pathogenic variants in CHD3. Furthermore, we add evidence of the importance of the application of massive parallel sequencing for NDD patients for whom the clinical diagnosis might be challenging and where deep phenotyping is extremely useful to accurately manage and follow up the patients.

Pulmonary arterial hypertension (PAH) is an infrequent disorder characterized by high blood pressure in the pulmonary arteries. It may lead to premature death or the requirement for lung and/or heart transplantation. Genetics plays an important and increasing role in the diagnosis of PAH. Here, we report seven additional patients with variants in SOX17 and a review of sixty previously described patients in the literature. Patients described in this study suffered with additional conditions including large septal defects, as described by other groups. Collectively, sixty-seven PAH patients have been reported so far with variants in SOX17, including missense and loss-of-function (LoF) variants. The majority of the loss-of-function variants found in SOX17 were detected in the last exon of the gene. Meanwhile, most missense variants were located within exon one, suggesting a probable tolerated change at the amino terminal part of the protein. In addition, we reported two idiopathic PAH patients presenting with the same variant previously detected in five patients by other studies, suggesting a possible hot spot. Research conducted on PAH associated with congenital heart disease (CHD) indicated that variants in SOX17 might be particularly prevalent in this subgroup, as two out of our seven additional patients presented with CHD. Further research is still necessary to clarify the precise association between the biological pathway of SOX17 and the development of PAH.

Hemophilia A is an X-linked disorder characterized by quantitative deficiency of coagulation factor VIII (FVIII) caused by pathogenic variants in the factor 8 (F8) gene. Our study’s primary objective was to identify genetic variants within the exonic region of F8 in 50 Colombian male participants with severe hemophilia A (HA). Whole-exome sequencing and bioinformatics analyses were performed, and bivariate analysis was used to evaluate the relationship between identified variants, disease severity, and inhibitor risk formation. Out of the 50 participants, 21 were found to have 17 different pathogenic F8 variants (var). It was found that 70% (var = 12) of them were premature truncation variants (nonsense, frameshift), 17.6% (var = 3) were missense mutations, and 11.7% (var = 2) were splice-site variants. Interestingly, 35% (var = 6) of the identified variants have not been previously reported in the literature. All patients with a history of positive inhibitors (n = 4) were found to have high-impact genetic variants (nonsense and frameshift). When investigating the relationship between variant location (heavy versus light chain) and specific inhibitor risk, 75% (n = 3) of the inhibitor participants were found to have variants located in the F8 light chain (p = 0.075), suggesting that conserved domains are associated with higher inhibitor risk. In summary, we identified genetic variants within the F8 that can possibly influence inhibitor development in Colombian patients with severe HA. Our results provide a basis for future studies and the development of further personalized treatment strategies in this population.

Summary: Background: Large language models (LLMs) are increasingly used in the medical field for diverse applications including differential diagnostic support. The estimated training data used to create LLMs such as the Generative Pretrained Transformer (GPT) predominantly consist of English-language texts, but LLMs could be used across the globe to support diagnostics if language barriers could be overcome. Initial pilot studies on the utility of LLMs for differential diagnosis in languages other than English have shown promise, but a large-scale assessment on the relative performance of these models in a variety of European and non-European languages on a comprehensive corpus of challenging rare-disease cases is lacking. Methods: We created 4967 clinical vignettes using structured data captured with Human Phenotype Ontology (HPO) terms with the Global Alliance for Genomics and Health (GA4GH) Phenopacket Schema. These clinical vignettes span a total of 378 distinct genetic diseases with 2618 associated phenotypic features. We used translations of the Human Phenotype Ontology together with language-specific templates to generate prompts in English, Chinese, Czech, Dutch, German, Italian, Japanese, Spanish, and Turkish. We applied GPT-4o, version gpt-4o-2024-08-06, to the task of delivering a ranked differential diagnosis using a zero-shot prompt. An ontology-based approach with the Mondo disease ontology was used to map synonyms and to map disease subtypes to clinical diagnoses in order to automate evaluation of LLM responses. Findings: For English, GPT-4o placed the correct diagnosis at the first rank 19·8% and within the top-3 ranks 27·0% of the time. In comparison, for the eight non-English languages tested here the correct diagnosis was placed at rank 1 between 16·9% and 20·5%, within top-3 between 25·3% and 27·7% of cases. Interpretation: The differential diagnostic performance of GPT-4o across a comprehensive corpus of rare-disease cases was consistent across the nine languages tested. This suggests that LLMs such as GPT-4o may have utility in non-English clinical settings. Funding: NHGRI 5U24HG011449 and 5RM1HG010860. P.N.R. was supported by a Professorship of the Alexander von Humboldt Foundation; P.L. was supported by a National Grant (PMP21/00063 ONTOPREC-ISCIII, Fondos FEDER).