BANKSY: A Spatial Omics Algorithm that Unifies Cell Type Clustering and Tissue Domain Segmentation
summary
Algorithm constructs a product space of cell and neighborhood transcriptomes by appending k‑nearest‑neighbor averaged expression to each cell, then runs standard single‑cell clustering or segmentation with a single tuneable weight lambda. Benchmarks on MERFISH, seq‑ and FISH‑based datasets show cleaner subtype separation, discovery of niche‑dependent cell states in mouse brain, successful QC/domain calling for organoids, and runtimes that are orders of magnitude faster than prior spatial methods.
AI-written summary · about
chapters (14)
- 0:00Institute and workshop intro
- 2:34Organizers' welcome
- 5:12Workshop schedule overview
- 7:52Speaker introduction
- 9:40Evolution of genomics
- 12:15Spatial omics challenges
- 14:47Current analysis limitations
- 20:23Introducing Banksy
- 26:19Banksy algorithm intuition
- 28:27Benchmarking and applications
- 33:51Slide-seq performance
- 36:10Tissue domain segmentation
- 38:26Scalability and summary
- 39:28Q&A session
related talks
29mUnveiling new maps of biology with Squidpy
27mClustering of high-content screen images to discover off-target phenotypes
29mHarnessing Multimodal Single-Cell Sequencing Data for Integrative Analysis
14mFrom Pixels to Pathways: Omni-Modal Cell Biology with AI
32mMulti-Omic Integration for Understanding Disease
27mKubernetes + AI Joining Forces in the Battle Against Cancer
31mHacking the Human Genome Using Clojure and Similarity Search
35mFrom cells to drug responses - machine learning in cancer research