Date |
2:00pm May 18, 2022 |
Speaker |
Keita Iida
Osaka University
|
Title |
Clustering single-cell and spatial transcriptomes through multifaceted biological aspects
|
Abstract
|
Single-cell transcriptomics have deepened our knowledge of biological
complexity in terms of molecular heterogeneity in cell populations.
However, conventional signature gene-based approaches may be
insufficient in capturing such complexity as genes can interact with
each other to regulate a certain biological function. Here, we
introduce ASURAT, a computational pipeline for simultaneously
performing unsupervised clustering and functional annotation of
disease, cell type, biological process, and signaling pathway activity
for single-cell transcriptomic data. Mimicking Saussure’s idea for
defining “meaning", we introduce correlation structures into
functionally categorized gene sets defined in knowledge-based
databases, such as Cell Ontology, Disease Ontology, Gene Ontology
databases, and Kyoto Encyclopedia of Genes and Genomes. Then, we apply
a novel correlation graph decomposition to each gene set, producing a
triplet we term "sign" containing biological description, subset of
genes, and correlation structure, which is the central concept. To
validate usability and clustering performance, we apply ASURAT on
single-cell RNA sequencing and spatial transcriptome datasets of human
pancreatic ductal adenocarcinoma (Moncada et al., Nat. Biotechnol. 38,
2020). We demonstrate that ASURAT is able not only to cluster spatial
transcriptomic data into functionally distinct tissue regions but also
identify de novo atypical regions.
|
|