Abstract
Cancer tissues are heterogeneous mixtures of tumour, stromal and immune cells, where each component comprises multiple distinct cell types and/or states. Mapping this heterogeneity and understanding the unique contributions of each cell type to the tumour transcriptome is crucial for advancing cancer biology, yet high-throughput expression profiles from tumour tissues only represent combined signals from all cellular sources. Computational deconvolution of these mixed signals has emerged as a powerful approach to dissect both cellular composition and cell-type-specific expression patterns. Here, we provide a comprehensive guide to transcriptomic deconvolution, specifically tailored for cancer researchers, presenting a systematic framework for selecting and applying deconvolution methods, considering the unique complexities of tumour tissues, data availability and method assumptions. We detail 43 deconvolution methods and outline how different approaches serve distinctive applications in cancer research: from understanding tumour-immune surveillance to identifying cancer subtypes, discovering prognostic biomarkers and characterizing spatial tumour architecture. By examining the capabilities and limitations of these methods, we highlight emerging trends and future directions, particularly in addressing tumour cell plasticity and dynamic cell states.
Citation
@article{dai2025guide,
title={A guide to transcriptomic deconvolution in cancer},
author={Dai, Yaoyi and Guo, Shuai and Pan, Yidan and Castignani, Carla and Montierth, Matthew D and Van Loo, Peter and Wang, Wenyi},
journal={Nature Reviews Cancer},
pages={1--20},
year={2025},
publisher={Nature Publishing Group UK London}
}