ESCAPED: Efficient Secure and Private Dot Product Framework for Kernel-based Machine Learning Algorithms with Applications in Healthcare

To train sophisticated machine learning models one usually needs many\ntraining samples. Especially in healthcare settings these samples can be very\nexpensive, meaning that one institution alone usually does not have enough on\nits own. Merging privacy-sensitive data from different sources is usually\nrestricted by data security and data protection measures. This can lead to\napproaches that reduce data quality by putting noise onto the variables (e.g.,\nin $\\epsilon$-differential privacy) or omitting certain values (e.g., for\n$k$-anonymity). Other measures based on cryptographic methods can lead to very\ntime-consuming computations, which is especially problematic for larger\nmulti-omics data. We address this problem by introducing ESCAPED, which stands\nfor Efficient SeCure And PrivatE Dot product framework, enabling the\ncomputation of the dot product of vectors from multiple sources on a\nthird-party, which later trains kernel-based machine learning algorithms, while\nneither sacrificing privacy nor adding noise. We evaluated our framework on\ndrug resistance prediction for HIV-infected people and multi-omics\ndimensionality reduction and clustering problems in precision medicine. In\nterms of execution time, our framework significantly outperforms the\nbest-fitting existing approaches without sacrificing the performance of the\nalgorithm. Even though we only show the benefit for kernel-based algorithms,\nour framework can open up new research opportunities for further machine\nlearning models that require the dot product of vectors from multiple sources.\n