Deconvolution Of Heterogeneous Tumor Samples Using Partial Reference Signals

Yufang Qin,Weiwei Zhang,Xiaoqiang Sun,Siwei Nan,Nana Wei,Hua-Jun Wu,Xiaoqi Zheng

DOI: https://doi.org/10.1371/journal.pcbi.1008452

2020-01-01

PLoS Computational Biology

Abstract:Deconvolution of heterogeneous bulk tumor samples into distinct cellular populations is an important yet challenging problem, particularly when only partial references are available. A common approach to dealing with this problem is to deconvolve the mixed signals using available references and leverage the remaining signal as a new cell component. However, as indicated in our simulation, such an approach tends to over-estimate the proportions of known cell types and fails to detect novel cell types. Here, we propose PREDE, a partial reference-based deconvolution method using an iterative non-negative matrix factorization algorithm. Our method is verified to be effective in estimating cell proportions and expression profiles of unknown cell types based on simulated datasets at a variety of parameter settings. Applying our method to TCGA tumor samples, we found that proportions of pure cancer cells better indicate different subtypes of tumor samples. We also detected several cell types for each cancer type whose proportions successfully predicted patient survival. Our method makes a significant contribution to deconvolution of heterogeneous tumor samples and could be widely applied to varieties of high throughput bulk data. PREDE is implemented in R and is freely available from GitHub (https://xiaoqizheng.github.io/PREDE).Author summaryTumor tissues are mixtures of different cell types. Identification and quantification of constitutional cell types within tumor tissues are important tasks in cancer research. The problem can be readily solved using regression-based methods if reference signals are available. But in most clinical applications, only partial references are available, which significantly reduces the deconvolution accuracy of the existing regression-based methods. In this paper, we propose a partial-reference based deconvolution model, PREDE, integrating the non-negative matrix factorization framework with an iterative optimization strategy. We conducted comprehensive evaluations for PREDE using both simulation and real data analyses, demonstrating better performance of our method than other existing methods.