Deep autoencoder enables interpretable tissue-adaptive deconvolution and cell-type-specific gene analysis

Single-cell RNA-seq has become a powerful tool for researchers to study biologically significant characteristics at explicitly high resolution, but its application on emerging data is currently limited by its intrinsic techniques. Here, we introduce TAPE, a deep learning method that connects bulk RNA-seq and single-cell RNA-seq to balance the demands of big data and precision. By taking advantage of constructing an interpretable decoder and training under a unique scheme, TAPE can predict cell-type fractions and cell-type-specific gene expression tissue-adaptively. Compared with existing methods on several benchmarking datasets, TAPE is more accurate (up to 40% performnace improvement on the real bulk data) and faster than the previous methods. It is sensitive enough to provide biologically meaningful predictions. For example, only TAPE can predict the tendency of increasing monocytes-to-lymphocytes (MLR) ratio in COVID-19 patients from mild to serious symptoms, whose estimated indices are consistent with laboratory data. More importantly, through the analysis of clinical data, TAPE shows its ability to predict cell-type-specific gene expression profiles with biological significance. Combining with single-sample gene set enrichment analysis (ssGSEA), TAPE also provides valuable clues for people to investigate the immune response in different virus-infected patients. We believe that TAPE will enable and accelerate the precise analysis of high-throughput clinical data in a wide range.