ELITE: Expression deconvoLution using lInear optimizaTion in bulk transcriptomics mixturEs

Understanding the cellular composition of tissue samples is crucial for identifying the molecular mechanisms underlying diseases and developing cellular targets for therapeutic interventions. Digital cytometry methods have been developed to predict tissue composition from bulk transcriptomic data, avoiding the high cost associated with single-cell profiling. Here, we present ELITE, a new digital cytometry method that utilizes linear programming to solve the deconvolution problem. ELITE uses as inputs a mixture matrix representing bulk measurements, and a signature matrix representing molecular fingerprints of the cell types to be identified. The signature matrix can be obtained from single-cell datasets or the literature, making ELITE more flexible than other methods that rely solely on single-cell data. We evaluated ELITE on three publicly available single-cell datasets and compared it with five other deconvolution methods, showing superior performance, particularly when there were cell types with similar expression profiles. As a case study, we evaluated the prediction of tumor cellularity using purity estimates from 20 different TCGA carcinoma datasets.