| Title: | GenoGraphNet: Pathway-Level Genomic Reasoning with Clinical Language Alignment for Idiopathic Male Infertility Prediction |
| Journal: | IEEE Access |
| Published: | 31 Jul 2026 |
| DOI: | https://doi.org/10.1109/access.2026.3718543 |
Publication 21321
WARNING: the interactive features of this website use CSS3, which your browser does not support. To use the full features of this website, please update your browser.
Abstract
Idiopathic male infertility (IMI) is genetically heterogeneous, and clinically relevant signals may be distributed across biological pathways rather than concentrated in single genes. We present GenoGraphNet, a pathway-level genomic reasoning framework with clinical language alignment for IMI risk modeling using UK Biobank exome-linked health records. The leakage-controlled analytical cohort contains 1,872 IMI cases and 10,608 controls, with a 15.0% case rate; precision-recall and positive-predictive-value-related quantities are therefore interpreted within this analytical cohort. GenoGraphNet projects rare functional variant burden from genes to pathways, performs graph message passing over pathway relatedness, encodes leakage-controlled pre-index clinical record-derived evidence with PubMedBERT, and aligns clinical evidence with pathway nodes for mechanism-aware multimodal fusion. The prediction protocol excludes label-defining infertility diagnoses, semen-analysis evidence used for case definition, infertility treatment records, assisted reproduction records, and post-index information from model inputs. In the leakage-controlled test set, GenoGraphNet achieved an AUROC of 0.892 (95% CI: 0.884-0.900), an AUPRC of 0.748 (0.731-0.765), an F1 score of 0.752 (0.736-0.767), a Brier score of 0.111 (0.106-0.116), and an expected calibration error of 0.029, outperforming clinical-genomic, text-only, genomics-only, and generic multimodal fusion baselines. Prioritized pathways were enriched for known infertility-related genes and spermatogenesis-related processes. These results suggest that leakage-controlled pathway-level genotype-phenotype alignment can support interpretable IMI risk modeling and hypothesis generation, but the model is not intended for direct clinical decision making. External validation in independent and diverse infertility cohorts remains required before translational use.</p>
8 Authors
- Wanqing Wang
- Zehua Lu
- Feihong Liu
- Chuyang Chen
- Yue Peng
- Wei Chen
- Chunyun Li
- Hongqiu Zhu
Enabling scientific discoveries that improve human health