Guidelines for cell-type heterogeneity quantification based on a comparative analysis of reference-free DNA methylation deconvolution software

• DOI: 10.1186/s12859-019-3307-2
• Publication Date: 2020-01-13T14:02:49Z
• Authors: Clémentine Decamps; Florian Privé; Raphael Bacher; Daniel Jost; Arthur Waguet; Eugene Andres Hou...; Eugene Lurie; Pavlo Lutsik; Aleksandar Milosa...; Michael Scherer; Michael G. B. Blum; Magali Richard
• Keywords: Initialization; Cell type

Abstract Background Cell-type heterogeneity of tumors is a key factor in tumor progression and response to chemotherapy. Tumor cell-type heterogeneity, defined as the proportion various cell-types tumor, can be inferred from DNA methylation surgical specimens. However, confounding factors known associate with values, such age sex, complicate accurate inference proportions. While reference-free algorithms have been developed infer proportions methylation, comparative evaluation performance these methods still lacking. Results Here we use simulations evaluate several computational pipelines based on software packages MeDeCom, EDec, RefFreeEWAS. We identify that accounting for confounders, feature selection, choice number estimated cell types are critical steps inferring find removal probes which correlated confounder variables reduces error by 30–35%, selection informative has similar effect. show Cattell’s rule scree plot powerful tool determine cell-types. Once pre-processing achieved, three deconvolution provide comparable results. observe all algorithms’ improves when inter-sample variation large or available samples large. under specific circumstances sensitive initialization method, suggesting averaging different solutions optimizing an avenue future research. Conclusion Based lessons learned, facilitate pipeline validation catalyze further improvement community, develop benchmark implement it R package medepir .