- Thursday, September 22, 2022
- 3:00 PM–4:00 PM
- NH 276
Mathematics & Statistics Colloquium welcomes David Zhang to lecture on "Super-Resolved Tissue Annotation in Spatial Transcriptomics."
Dr. David Zhang joins us from the University of Pennsylvania for this lecture on Thursday, September 22. Join us for refreshments in the Reading Room at 3:00, and then in NH 276 at 3:10 for the lecture.
Abstract:
Annotation of tissue sections are crucial to the diagnosis and study of cancer. Manual annotation by pathologists is a labor-intensive process and may vary greatly between experts. We propose a data-drive approach that automatically annotates tissues at the pixel level based on histology images and spatial transcriptomics (ST). Using a pretrained hierarchical image pyramid transformer (HIPT) model, we compute histology embeddings in a training-free manner. Histology embeddings are then utilized to perform unsupervised segmentation of the tissue section. Moreover, we increase the spatial resolution of the paired ST data from spot-level to pixel-level based on the distance defined by histology features. Differential expression analysis are then conducted to select the top most over-expressed genes in every tissue cluster, which in turn are mapped to biological functions using external databases to provide functional annotations of the cluster. Compared to existing methods, our method does not require large training datasets or high-resolution ST measurements. We demonstrate the effectiveness of our method by applying it to breast cancer tissues.
Location details
North Hall 276
