Federated Learning Architectures for Privacy-Preserving Pharmacogenomic Databases

Karan Radadiya

---

ABSTRACT

This study examines the opportunities of federated learning as a privacy-friendly approach for conducting pharmaceutical research using a pharmacogenomic database. Pharmacogenomic data are highly sensitive and distributed across numerous institutions, making safe data sharing a significant challenge. The research evaluated anonymized pharmacogenomic data from 5,000 records across 3 drug companies and 4 hospitals using federated learning systems in a research environment. They were used to evaluate model performance by measuring accuracy and F1-score, performing 10-fold cross-validation, and conducting t-tests and ANOVA. Findings showed that federated models achieved prediction accuracies of 85-90%, a F1-score of 0.88, a mean cross-validation score of 90%, and were statistically significant at p < 0.05. The findings also show that federated learning can promote privacy, support distributed data analysis, and enable collaborative research in pharmacogenomics, and that sharing raw data is not a prerequisite across institutions. Federated learning is a promising technology for effective and safe pharmacogenomic predictive analytics, and its use in drug discovery and personalized medicine raises issues of model synchronization, infrastructure, and large-scale implementation challenges.

Keywords: Federated Learning, Privacy-Preserving, Pharmacogenomics, Personalized Medicine, Drug Discovery, Data Security

How to cite this article: Radadiya K. Federated Learning Architectures for Privacy-Preserving Pharmacogenomic Databases. Int J Drug Deliv Technol. 2026;16(20s): 99. DOI: 10.25258/ijddt.16.20s.12

Source of support: Nil.

Conflict of interest: None