题    目：Advancing Intelligent Spatial Transcriptomics by Deep Generative Models

主讲人：张驰浩

单    位：中国科学院

时    间：2024年8月19日 10:00

腾讯ID：598 617 222

摘    要：Spatial transcriptomics, which profiles gene expression with precise spatial information, offers an unprecedented opportunity to investigate complex biological systems. This presentation will highlight the role of deep generative models in enhancing our ability to analyze and interpret spatial transcriptomic data. We introduce STAMarker, a deep learning framework that identifies spatially variable genes with greater precision, especially in sparse datasets. By leveraging graph-attention autoencoders and multilayer perceptron classifiers, STAMarker reveals genes that are pivotal to understanding spatial domains within tissues. STADiffuser, another deep generative model, is presented as a novel tool for simulating spatial transcriptomic data. Its integration of an autoencoder with a graph attention mechanism and a latent diffusion model allows for the generation of high-fidelity data, enabling advanced analyses such as imputation, super-resolution, and 3D slice imputation. This talk will showcase how these models are transforming spatial transcriptomics analysis, paving the way for deeper insights into cellular systems and tissue organization.

简    介：张驰浩，中国科学院数学与系统科学研究院，助理研究员，曾前往美国加州大学洛杉矶分校访问和日本东京大学开展博士后工作。主要研究方向为机器学习与生物信息。近年来以通讯或第一作者在IEEE TPAMI、JMLR、PR、NAR、IEEE TKDE等国际著名期刊上发表多篇论文。