It is becoming clear that bulk gene expression measurements represent an average over very different cells. Elucidating the expression and abundance of each of the encompassed cells is key to disease understanding and precision medicine approaches. A first step in any such deconvolution is the inference of cell type abundances in the given mixture. To combat this task, we develop DECODE, a deep unfolded non-negative matrix factorization technique. We show that our method outperforms previous approaches on a range of synthetic and real data sets.

UNMET NEED
Cell type composition can inform disease subtyping, prognosis and therapy. However, experimental measurement of the abundance of specific cell types in a sample is laborious and costly. Computational deconvolution has shown promising results but is limited by scarce training data and inaccurate expression profile information for specific cells.

OUR SOLUTION
DECODE is a deep learning framework for cell type deconvolution from bulk expression data. It builds on a unique deep unfolded non-negative matrix factorization technique and has been shown to be superior to other computational solutions thanks to its innovative architecture and training procedure and its flexibility in considering prior cell-type-specific expression profile information in the deconvolution process.

APPLICATIONS
Specific cell types have been associated with improved clinical outcomes and response to immunotherapy, hence the deconvolution of tumors to their constituent cell types has important therapeutic applications.

STATUS
A manuscript is in revision to Digital Medicine and Healthcare Technology journal.

INTELLECTUAL PROPERTY
Provisional application