Published June 6, 2025 | Version v3

4917 Phenopackets and Prompts in 10 Languages for LLM Evaluation, with results for GPT-4o and Meditron3-70B

  • 1. ROR icon Berlin Institute of Health at Charité - Universitätsmedizin Berlin
  • 2. ROR icon Lawrence Berkeley National Laboratory
  • 3. The Jackson Laboratory for Genomic Medicine

Description

Collection of 4917 GA4GH Phenopackets and of prompts generated from them in 10 languages:
English, Chinese, Czech, Dutch, French, German, Italian, Japanese, Spanish, and Turkish.


Used in project about multilingual capability of LLMs (GPT-4o and Meditron3-70B) in rare genetic disease diagnostics. Also contains our results of the comparison of the 10 languages for each of the models. Paper published in eBioMedicine here.

Phenopacket Store version 0.1.19. Phenopackets found here are the subset for which all HPO terms are known in the 10 languages, using HPO version 2024-12-12 

https://github.com/p2gx/phenopacket2prompt

https://github.com/monarch-initiative/pheval.llm

Files

4917promptsX10langs.zip

Files (76.9 MB)

Name Size Download all
md5:758f233ee505305c8c6165da5a7dbaa5
59.6 MB Preview Download
md5:1247e8c377091b6019d676680adce7e1
8.9 MB Preview Download
md5:d9313ccfa48b7d9cdd5fbef660886e45
200.6 kB Preview Download
md5:6a88131540ea3e5dd00b342e96996f4b
8.2 MB Preview Download

Additional details

Related works

Is supplemented by
Dataset: 10.5281/zenodo.15065293 (DOI)