Functional genomics data: privacy risk assessment and technological mitigation

PPML-Omics: a Privacy-Preserving federated Machine Learning method protects patients’ privacy in omic data

“…Bulk RNA-seq can directly disclose patients’ gene expression, while machine learning models trained with bulk RNA-seq can indirectly leak patients’ signature expressed genes [43]. Here, we collected data from TCGA ( Methods and Supplementary Table 1 ) and compared (i) computing resource requirements, (ii) privacy-preserving capabilities, and (iii) utility on data of different methods by working on the cancer classification task with gene expression as inputs.…”

“…However, those methods introduce different trade-offs compared to DP-based solutions, such as bigger computation and communication overheads, which might be difficult to satisfy by some clients in practice, while DP yields a privacy-utility trade-off, under which we could protect privacy by sacrificing some data utility. Currently, many works in the literature utilized the DP notion to provide formal privacy guarantees for the participants in the research of SNPs and GWAS, which is a relatively narrow and specific problem in genomic studies and whose data is obtained by post-processing the raw sequencing data [9,[38][39][40][41][42][43][44] (Supplementary Section 1.3). In addition to SNPs, raw sequencing data saved in matrix format contains much more sensitive information.…”

Routine omics collection is a golden opportunity for European human research in space and analog environments

“…However, the identifiability of other omics types, such as transcriptomics and proteomics, is also an emerging area of research. 113 , 114 , 115 , 116 Furthermore, the generation of multiomics datasets adds an additional layer of complexity because there could be relationships between the different omic types that increase the overall likelihood of identification. 117 Additionally, raw data hold increased potential for identification compared with processed data, so file formats are an important consideration for appropriate data dissemination.…”

Computational methods and translational applications for targeted next‐generation sequencing platforms

“…59 Seamless integration of omics data with information from the EHR and additional variables of clinical relevance such as dietary habits, levels of physical activity or vital signs (body temperature, pulse rate, blood pressure) that can nowadays be monitored efficiently through portable and wearable devices has the potential to revolutionize the emerging digital health ecosystem. Of course, this evolving scenario will also bring forth novel privacy challenges and robust approaches to avoid genotypic and phenotypic information leakage, 89 which may be particularly troubling when germline data is involved. 90 From a translational perspective, the continued success of NGS platforms will largely depend on their ability to demonstrate clinical utility and optimize treatment choices.…”