Growing applications of deep learning on sensitive genomics and biomedical data introduce challenging privacy and secure problems. Homomorphic encryption (HE) is one of appropriate cryptographic techniques to provide secure machine learning evaluation by directly computing over encrypted data, so that allows the data owner and model owner to outsource processing of sensitive data to an untrusted server without leaking any information about the data. However, most current HE schemes only support limited arithmetic operations, which significantly hinder their applications to support secure deep learning algorithm. Considering the potential performance loss introduced for approximating activation function, in this paper, we develop a novel HE friendly deep network, named Residue Activation Network (ResActNet) to implement precise privacy-preserving machine learning algorithm with a non-approximating activation on HE scheme. We considered a residue activation strategy with a scaled power activation function in the deep network. In particular, a scaled power activation (SPA) function is set within the HE scheme, and so that can be directly deployed on HE computation. Moreover, we proposed a residue activation strategy to constrain the latent space in the training process for alleviating the optimization difficulty. We comprehensively evaluate ResActNet using diverse genomics datasets and widely-used image datasets. Our results demonstrate that ResActNet outperforms other alternative solutions to secure machine learning with HE and achieves low approximation errors in classification and regression tasks.

Prostate cancer is one of the deadliest cancers worldwide. An accurate prediction of pathological stages using the expressions and interactions of genes is effective for clinical assessment and treatment. However, identification of interactions using biological procedure is time consuming and prohibitively expensive. A graph is a powerful representation for the complex interactome of genes, their transcripts, and proteins. Recently, Graph Neural Networks (GNNs) have gained great attention in machine learning due to their capability to capture the graphical interactions among data entities. To leverage GNNs for predicting pathological stage stages, we developed an end-to-end graph representation and learning model, namely E2EGraph, which can automatically generate a graph representation using gene expression data and a multi-head graph attention network to learn the strength of interactions among genes and make the prediction. To ensure the reliability of model prediction, we identify critical components of graph representation and GNN model to interpret prediction results from multiple perspectives at gene and patient levels. We evaluated E2EGraph to predict pathological stages of prostate cancer using The Cancer Genome Atlas (TCGA) data. Our experimental results demonstrate that E2EGraph reaches the state-of-art prediction performance while being effective in identifying marker genes indicated by interpretability. Our results point to a direction where adaptive graph construction and attention based GNNs can be leveraged for various prediction tasks and interpretation of model prediction in a variety of data domains including disease prediction.

Today, the rapid development of deep learning has spread across all walks of life, and it can be seen in various fields such as image classification, automatic driving, and medical imaging diagnosis. Convolution Neural Networks (CNNs) are also widely used by the public as tools for deep learning. In real life, if local customers implement large-scale model inference first, they need to upload local data to the cloud, which will cause problems such as data leakage and privacy disclosure. To solve this problem, we propose a framework using homomorphic encryption technology. Our framework has made improvements to the batch operation and reduced the complexity of layer connection. In addition, we provide a new perspective to deal with the impact of the noise caused by the homomorphic encryption scheme on the accuracy during the inference. In our scheme, users preprocess the images locally and then send them to the cloud for encrypted inference without worrying about privacy leakage during the inference process. Experiments show that our proposed scheme is safe and efficient, which provides a safe solution for users who cannot process data locally.