SCIM: universal single-cell matching with unpaired feature sets
Autoencoder
DOI:
10.1093/bioinformatics/btaa843
Publication Date:
2020-09-14T11:44:22Z
AUTHORS (127)
ABSTRACT
Recent technological advances have led to an increase in the production and availability of single-cell data. The ability integrate a set multi-technology measurements would allow identification biologically or clinically meaningful observations through unification perspectives afforded by each technology. In most cases, however, profiling technologies consume used cells thus pairwise correspondences between datasets are lost. Due sheer size can acquire, scalable algorithms that able universally match carried out one cell its corresponding sibling another technology needed.We propose Single-Cell data Integration via Matching (SCIM), approach recover such two more technologies. SCIM assumes share common (low-dimensional) underlying structure distribution is approximately constant across It constructs technology-invariant latent space using autoencoder framework with adversarial objective. Multi-modal integrated pairing bipartite matching scheme operates on low-dimensional representations. We evaluate simulated cellular branching process show cell-to-cell matches derived reflect same pseudotime dataset. Moreover, we apply our method real-world scenarios, melanoma tumor sample human bone marrow sample, where pair from scRNA dataset their CyTOF achieving 90% 78% cell-matching accuracy for samples, respectively.https://github.com/ratschlab/scim.Supplementary available at Bioinformatics online.
SUPPLEMENTAL MATERIAL
Coming soon ....
REFERENCES (43)
CITATIONS (50)
EXTERNAL LINKS
PlumX Metrics
RECOMMENDATIONS
FAIR ASSESSMENT
Coming soon ....
JUPYTER LAB
Coming soon ....