Case Study: RIKEN achieves CNN-based feature extraction on non-image genomic data with MathWorks (MATLAB & Deep Learning Toolbox)

RIKEN Develops a Method to Apply CNN to Non-Image Data

RIKEN, Japan’s largest basic and applied science research organization, faced the challenge of detecting small phenotype and genomic variations from non-image data such as RNA sequences—tasks that make it difficult to identify relevant genes or features and to apply convolutional neural networks (CNNs) directly. In their Deepinsight project, RIKEN worked with MathWorks tools—using MATLAB and the Deep Learning Toolbox—to convert non-image data into image format and prepare it for CNN-based analysis.

Using MathWorks’ MATLAB and Deep Learning Toolbox, RIKEN clustered similar elements, converted sequence data into images, and applied CNNs (leveraging multiple GPUs) to perform feature extraction and classification. The MathWorks-based solution enabled quick feature discovery, helped identify hidden biological mechanisms, and significantly reduced coding time while supporting GPU-accelerated training and inference.

RIKEN