In this article, we review two challenging computational questions in protein science: neoantigen prediction and protein structure prediction. Both topics have seen significant leaps forward by deep learning within the past five years, which immediately unlocked new developments of drugs and immunotherapies. We show that deep learning models offer unique advantages, such as representation learning and multi-layer architecture, which make them an ideal choice to leverage a huge amount of protein sequence and structure data to address those two problems. We also discuss the impact and future possibilities enabled by those two applications, especially how the data-driven approach by deep learning shall accelerate the progress towards personalized biomedicine.

In data outsourcing, a client stores a large amount of data on an untrusted server; subsequently, the client can request the server to compute a function on any subset of the data. This setting naturally leads to two security requirements: confidentiality of input data, and authenticity of computations. Existing approaches that satisfy both requirements simultaneously are built on fully homomorphic encryption, which involves expensive computation on the server and client and hence is impractical. In this paper, we propose two verifiable homomorphic encryption schemes that do not rely on fully homomorphic encryption. The first is a simple and efficient scheme for linear functions. The second scheme supports the class of multivariate quadratic functions, by combining the Paillier cryptosystem with a new homomorphic message authentication code (MAC) scheme. Through formal security analysis, we show that the schemes are semantically secure and unforgeable.

Read mapping is a key task in next-generation sequencing (NGS) data analysis. To achieve an optimal combination of accuracy, speed, and low memory footprint, popular mapping tools often focus on identifying one or a few best mapping locations for each read. However, for many downstream analyses such as prediction of genomic variants or protein binding motifs located in repeat regions, isoform expression quantification, metagenomics analysis, it is more desirable to have a comprehensive set of all possible mapping locations of NGS reads. In this paper, we introduce AMAS, a read mapping tool that exhaustively searches for possible mapping locations of NGS reads in a reference sequence within a given edit distance. AMAS features improvements of the mapping, partition, and filtration of adaptive seeds to speed up the read mapping. Performance results on simulated and real data sets show that AMAS run several times faster than other state-of-the-art read mappers while achieving similar sensitivity and accuracy. AMAS is implemented in C++ and is freely available at https://sourceforge.net/projects/ngsamas/.