Sim, J. J., & Zhou, W. (n.d.). An End-to-end Privacy-preserving Framework for Predicting Cancer Outcomes.
Abstract:
Genomic factors are strongly associated with cancer treatment outcomes and survival rates of patients. Although such factors have traditionally been extracted via expert knowledge and annotations, the advancements in artificial intelligence have led to an increasing focus on automated feature extraction based on training on large amounts of genomic data. However, genomic data is widely considered as sensitive, as their missuse can lead to dire ramifications such as social stigmatization, limited employment prospects, and loaded insurance policies. For instance, it has been shown that reidentification of patients can be achieved with a set of less than 80 independent single nucleotide polymorphisms (SNPs). As such, researchers often face difficulties in performing analysis across national borders, which relies on data access agreements that are often cumbersome to navigate.
Precision medicine is fundamentally a data-driven endeavour which relies on the availability of large amounts of multi-modal data, and the challenges in securing access to relevant datasets hinders the advancement of the field. In this work, we address the privacy concerns regarding the sharing of sensitive genomic data, by demonstrating a privacy-preserving framework to use models securely.
License type:
Publisher Copyright
Funding Info:
This research / project is supported by the A*STAR - Digital Trust Centre (DTC) Research Grant
Grant Reference no. : DTC-RGC-01