Deep-learning augmented RNA-seq analysis of transcript splicing

• Published in: Nature methods Vol. 16; no. 4; pp. 307 - 310
• Main Authors: Zhang, Zijun, Pan, Zhicheng, Ying, Yi, Xie, Zhijie, Adhikari, Samir, Phillips, John, Carstens, Russ P., Black, Douglas L., Wu, Yingnian, Xing, Yi
• Format: Journal Article
• Language: English
• Published: New York Nature Publishing Group US 01.04.2019; Nature Publishing Group
• Subjects: 631/114/1305; 631/114/2415; 631/114/794; 631/337/1645/1792; 631/61/212/2019; Algorithms; Alternative Splicing; Artificial neural networks; Bayes Theorem; Bayesian analysis; Big Data; Biochemistry; Bioinformatics; Biological Microscopy; Biological properties; Biological samples; Biological Techniques; Biomedical and Life Sciences; Biomedical Engineering/Biotechnology; Brief Communication; Computer applications; Data management; Datasets; Deep Learning; Empirical analysis; Epigenomics; Exons; Gene expression; Gene Expression Profiling; Gene Expression Regulation; Gene sequencing; Hep G2 Cells; High-Throughput Nucleotide Sequencing; Humans; Hypotheses; Hypothesis testing; K562 Cells; Knowledge bases (artificial intelligence); Life Sciences; Machine learning; Medical research; Models, Statistical; Neural networks; Proteomics; Public domain; Reproducibility of Results; Ribonucleic acid; RNA; RNA - analysis; RNA - genetics; RNA sequencing; RNA Splicing; Sequence analysis; Sequence Analysis, RNA; Signal Processing, Computer-Assisted; Splicing; Transcription
• ISSN: 1548-7091; 1548-7105; 1548-7105
• DOI: 10.1038/s41592-019-0351-9

A major limitation of RNA sequencing (RNA-seq) analysis of alternative splicing is its reliance on high sequencing coverage. We report DARTS ( https://github.com/Xinglab/DARTS ), a computational framework that integrates deep-learning-based predictions with empirical RNA-seq evidence to infer differential alternative splicing between biological samples. DARTS leverages public RNA-seq big data to provide a knowledge base of splicing regulation via deep learning, thereby helping researchers better characterize alternative splicing using RNA-seq datasets even with modest coverage. DARTS first uses public domain data to train a deep neural network to predict differential alternative splicing; the predictions are then combined with observed RNA-seq data in a Bayesian framework to infer changes in alternative splicing between biological samples.