Semi-supervised Nonnegative Matrix Factorization for gene expression deconvolution: A case study

► Sample heterogeneity is an issue inherent in many gene expression studies. ► Computational deconvolution is an attractive alternative to laboratory techniques. ► Unsupervised deconvolution methods cannot easily extract relevant signatures. ► We propose a semi-supervised approach that uses markers to improve estimation. ► Application to a real dataset showed its potential to provide more meaningful results.