Fig. 1. PPML-Omics: A privacy-preserving federated machine learning method protects patients’ privacy in omic data.

(A) Schematic overview of the relationships and interactions between distributed data owners, aggregators, attackers, and techniques in the field of secure and private AI. (B) Schematic overview of different methods, including centrally trained method, federated learning (FL), FL with differential privacy (DP), and PPML-Omics. (C) Illustration of three representative tasks, datasets, and attacks of omic data here for demonstrating the utility and privacy-preserving capability of PPML-Omics, including the (i) cancer classification with bulk RNA-seq, (ii) clustering with scRNA-seq, and (iii) integration of morphology and gene expression with spatial transcriptomic.