gms | German Medical Science

Gene-expression profiling of bulk tumor tissue facilitates the detection of gene regulation in tumor cells. However, differential gene expression can originate from both tumor cells and the cellular composition of the surrounding tissue. The cellular composition is not accessible in bulk sequencing but can be estimated computationally.

We propose Digital Tissue Deconvolution (DTD) to estimate cellular compositions from bulk sequencing data. Formally, DTD addresses the following problem: Given the expression profile y of a tissue, what is the cellular composition c of that tissue? If X is a matrix whose columns are reference profiles of individual cell types, the composition c can be computed by minimizing L(y-Xc) for a given loss function L. Current methods use predefined all-purpose loss functions. They successfully quantify the dominating cells of a tissue, while often falling short in detecting small cell populations.

In Görtler et al. [1], we presented a method to adapt the loss function to the deconvolution problem of interest. Here, we introduce the related R package "DTD". It provides all implementations, functions and routines for loss-function learning. Visualization functions are included to assess the quality of a deconvolution model and to gain additional information on the loss-function learning procedure. We present our package in an exemplary analysis, where we estimate immune cell quantities from gene-expression profiles of melanoma specimens. Using loss-function learning we increased the accuracy from a correlation of 29% to 72% between true and estimated cellular proportions.

Die Autoren geben an, dass kein Interessenkonflikt besteht.

Die Autoren geben an, dass kein Ethikvotum erforderlich ist.

Der Beitrag wurde bereits publiziert: Workshop on Computational Models in Biology and Medicine 2019, Braunschweig