wnchang/ictd repository overview

We developed a semi-supervised deconvolution method for tissue transcriptomics data that addressed the key challenges in existing methods, and we consider it has the following novelties: (1) a novel mathematical condition of cell type “identifiability” that grants the potential to detect novel cell and sub cell types without pre-specifying the cell types; (2) a semi-supervised knowledge transfer that is highly robust by handling the variations from independent single or bulk cell training data measured by diverse platforms, (3) a constrained non-negative matrix factorization (NMF) model to effectively handle the co-linearity among cell proportions; (4) a local low rank screening approach to identify cell type specific functional variations. Notably, ICTD uses a data driven approach to extract TME-specific cell type signature genes as information basis to infer and annotate cell types, hence it enables the application to a variety of tissue microenvironments, including cancer, inflammatory disease, blood and hematopoietic system, and brain of human and mouse.

Github repository: https://github.com/zy26/ICTD⁠

Web server: https://ictd.ccbb.iupui.edu⁠