Abstract: Transcriptomic analyses have increasingly contributed to our understanding of the intricate biological processes involved in the emergence of auto-immune diseases or tumour-promoting environments. However, classical bulk analyses ignore the intrinsic complexity of  biological samples, by averaging measurements over multiple distinct cell populations. It is therefore unclear whether a change in the gene expression between samples results from a variation of the cell type proportions or from a biological factor (Shen-Orr and Gaujoux, 2013).
To remove this ambiguity, deconvolution algorithms can estimate the proportions of cell populations from a bulk transcriptome using the reference transcriptome of purified cell populations. Traditionally, most approaches, including the gold standard CIBERSORT algorithm (Abbas et al., 2009), retrieve the cell proportions of a mixture assuming the linear assumption that each gene expression is the sum of each cell population’s contribution weighted by their corresponding relative frequency in the sample. However, none of these methods account for the transcriptomic covariance structure and address the crucial problem of co-transcriptomic expression between the genes. The first goal of our project aims at studying the impact of highly correlated structures assuming a sparse structure learnt by using the gLasso algorithm (Friedman, Hastie, and Tibshirani, 2008) on the performances of the canonical deconvolution algorithms using a reference-based method. Then, we will develop a new deconvolution method that integrates both the average expression and the covariance structure of the reference transcriptomic profiles to estimate cellular ratios, resolving noise effect induced by the interaction terms between gene transcripts.

Keywords: Deconvolution; Covariance; gLasso; Transcriptomic; Cell Population

Abbas, Alexander R et al. (2009). Deconvolution of Blood Microarray Data Identifies Cellular Activation Patterns in Systemic Lupus Erythematosus. PloS One, 4 (7): e6098.
Friedman, Jerome, Trevor Hastie, and Robert Tibshirani (2008). Sparse Inverse Covariance Estimation with the Graphical Lasso.. Biostatistics (Oxford, England) 9 (3): 432–41.
Shen-Orr, Shai S., and Renaud Gaujoux. (2013). Computational Deconvolution: Extracting Cell Type-Specific Information from Heterogeneous Samples. Current Opinion in Immunology 25 (5): 571–78.