Deep semi-supervised consistency regularization for accurate cell type fraction and gene expression estimation

Cell deconvolution is the estimation of cell type fractions and cell type-specific gene expression from mixed data. A challenge in cell deconvolution is the scarcity of realistic training data and the strong domain shift observed in synthetic training data. We show that consistency regularization in two deep neural networks significantly improve deconvolution performance. Our framework, DISSECT, achieves state-of-the-art performance on both deconvolution tasks, outperforming competing algorithms across several gene expression datasets by up to 14 percentage points. DISSECT can be adapted to deconvolve other biomedical data types, as exemplified by our proteomics deconvolution experiments.