Towards Reproducible and Privacy-preserving Analyses Across Federated Repositories for Omics data

Abstract

Even when duly anonymized, health research data has the potential to be disclosive and there- fore requires special safeguards according to the European General Data Protection Regulation (GDPR). Furthermore, the incorporation of FAIR principles (Findable, Accessible, Interoperable, Reusable) for a more favorable reuse of existing data, calls for an approach where sensitive data is kept locally and only metadata and aggregated results are shared. Additionally, since central pool- ing is discouraged by ethical, legal, and societal issues, it is more frequent to observe maturing data management frameworks, and platforms adopting the federated approach. Current implementations of privacy-preserving analysis frameworks seem to be limited when data becomes very large (millions of rows, hundreds of variables). Biological samples data, col- lected by high-throughput technologies, such as Next Generation Sequencing (NGS), which allows to sequence entire genomes, are examples of this kind of data. The term "genomics" refers to the field of science that studies genomes. The Omics tech- nologies intend to produce a systematic identification of all mRNA (transcriptomics), proteins (proteomics), and metabolites (metabolomics), respectively, present in a given biological sample. In the particular case of Omics data, these data are produced by computational workflows known as bioinformatics pipelines. The reproducibility of these pipelines is hard and it is often underestimated. Nevertheless, it is important to generate trust in scientific results, and therefore, is fundamental to know how these Omics data were generated or obtained. This work will leverage on the promising results of current open-source implementations for distributed privacy-preserving analyses, while aiming at generalizing the approach and addressing some of their shortcomings. To enable the privacy-preserving analysis of Omics data, we introduced the "resource" con- cept, implemented in one of the studied solutions. The results were promising, seeing that the privacy-preserving analysis was effective when us- ing the DataSHIELD framework in conjunction with the "resource R" package. We also concluded that the adoption of specialized DataSHIELD packages for Omics analyses is a viable pathway to leverage the privacy-preserving for Omics data. To address the reproducibility challenges, we defined a database model to represent the steps, commands and operations executed by the bioinformatics pipelines. The database model is promising, but to accomplish all reproducibility requirements, including container support and integration with code sharing platforms, it is necessary to use other tools, such as Nextflow or Snakemake, with dozens of other tested and mature functions.