E1E2E3:EnG1G2G3:GnIn our laboratory, we aim to understand disease dynamics and immunity 53scRNA-seq dataGenesTransformed imageExpressionTransform to image with DeepInsightDeep learningImage input, learn, feature extractionIdentified cell typesPrediction (identification)at the molecular level to conquer diseases such as cancer. Effective utiliza-tion of rapidly developing omic profiling technologies and the introduction of personalized/precision/preventive medicine have become major goals of medi-cal research, shifting away from traditional approaches that do not adequately consider the individuality of each patient. Our laboratory develops strategies to address these challenges by bringing the ideas and methods from mathematics and computational sciences to the medical domain. Recently, we proposed a new paradigm for deep learning to effectively handle sparse and interconnected data to discover biologically relevant features (DeepInsight / DeepFeature / DeepInsight-3D / scDeepInsight). We developed a unique technique, “Dee-pInsight”, which transforms omics data to image-like formats for subsequent use with convolutional neural networks; using this technique we were able to distinguish cancer types. We further developed a technique, “DeepFeature”, to analyze the inner workings of the neural network and identify how it discrimi-nates among cancers. Through this system, we discovered a new signaling sys-tem that can differentiate the characteristics of individual cancer types, showing that deep learning can be used to make new scientific discoveries. Furthermore, we have developed an advanced version, “DeepInsight-3D”, which uses multi-omics data, gene mutations, gene expression, and copy number alterations to predict drug efficacy. Lastly, based on our DeepInsight technique, we developed “scDeepInsight”, a method to identify cell types from single-cell RNA sequenc-ing data. It achieved an accuracy 7% greater than that of other cell-type identi-fication methods, helping to better understand complex biological systems and various diseases.Figure: Overview of the scDeepInsight methodJia S, Lysenko A, Boroevich KA, Sharma A, Tsunoda T. scDeepInsight: a supervised cell-type identification method for scRNA-seq data with deep learning. Brief Bioinform 24, bbad266 (2023)Recent Major PublicationsJia S, Lysenko A, Boroevich KA, Sharma A, Tsunoda T. scDeepInsight: a supervised cell-type identification method for scRNA-seq data with deep learning. Brief Bioinform 24, bbad266 (2023)Hamba Y, Kamatani T, Miya F, Boroevich KA, Tsunoda T. Topologically associating domain underlies tissue specific expression of long intergenic noncoding RNAs. iScience 26, 106640 (2023)Sharma A, Lysenko A, Boroevich KA, Tsunoda T. Deep-Insight-3D architecture for anti-cancer drug response prediction with deep-learning on multi-omics. Scien-tific Reports 13, 2483 (2023)Invited presentationsTsunoda T. “Deep learning pioneers omics medical sci-ence” Special workshop – Artificial Intelligence (AI) Research in the Post-Genome Era, The 61st Annual Meeting of Japan Society of Clinical Oncology (Yoko-hama, Japan) October 2023Tsunoda T. “Deep learning and mathematics pioneer new omics medical science” Symposium 2 – Omics Fron-tiers, The 44th Annual Meeting of the Japanese Society of Inflammation and Regeneration (Osaka, Japan) July 2023Laboratory for Medical Science MathematicsTeam Leader: Tatsuhiko Tsunoda
元のページ ../index.html#59