PESTO: prior evidence for gene-phenotype associations (code, results and model answers, v0.3.3)

Abstract

PESTO (Prior Evidence Status from Text and Ontologies) classifies the prior evidence for a gene–phenotype association as Novel, Hypothesized, Existing or Established, from the PubMed literature read by a language model and, separately, from the Open Targets Platform. This record accompanies the manuscript Large language model classifies prior evidence in gene–phenotype associations (Guez et al., 2026). It contains: PESTO-release-v0.3.3.zip: the code (Python package pesto-genetics 0.3.3, also on PyPI) and the scripts and tables that redraw every figure and table of the manuscript (github.com/jeremyguez/PESTO-release). pesto_paper_runs.tar.gz: every saved model answer behind the published figures (one JSON per gene–phenotype pair, with the retrieved corpus, the prompts and the response) and the Open Targets match records. pesto_paper_inputs.tar.gz: inputs and intermediate tables of the scripts that drew the benchmark pairs and queried the models. Both archives extract into paper/ of the code repository (paper/fetch_data.sh does it). Code is under the MIT licence. The tables and model answers produced for this work are under CC-BY 4.0; third-party content they contain (PubMed abstracts, GenCC, GO, EFO, gene lists) keeps the terms of its source.