Robust and reproducible classification of rare cellular subsets/signatures (RCS) in single-cell technologies within a transfer learning framework

Rare cellular subsets are a marginally infrequent set of cells, represented by a transcriptional signature or an immunophenotype- notype. The importance of RCS clinically in various clinical domains, from oncology to infectious diseases, has propelled research to define clinically actionable signatures. In this dissertation, we asked if there is a robust and reproducible computational method to classify RCS, improving the downstream hypothesis testing for research or clinical needs. In this investigation, we address two current limitations of computationally classifying cellular subsets: Transcriptionally, how well can RCS be classified? And how well does a classifier generalize across datasets, even if the sources/platforms differ?