Skip to main page content
U.S. flag
See More

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2014 Mar 1;30(5):682-9.
doi: 10.1093/bioinformatics/btt566. Epub 2013 Oct 1.

MMAD: microarray microdissection with analysis of differences is a computational tool for deconvoluting cell type-specific contributions from tissue samples

David A Liebner  1 , Kun Huang, Jeffrey D Parvin
Affiliations

MMAD: microarray microdissection with analysis of differences is a computational tool for deconvoluting cell type-specific contributions from tissue samples

David A Liebner et al. Bioinformatics. .

Abstract

Background: One of the significant obstacles in the development of clinically relevant microarray-derived biomarkers and classifiers is tissue heterogeneity. Physical cell separation techniques, such as cell sorting and laser-capture microdissection, can enrich samples for cell types of interest, but are costly, labor intensive and can limit investigation of important interactions between different cell types.

Results: We developed a new computational approach, called microarray microdissection with analysis of differences (MMAD), which performs microdissection in silico. Notably, MMAD (i) allows for simultaneous estimation of cell fractions and gene expression profiles of contributing cell types, (ii) adjusts for microarray normalization bias, (iii) uses the corrected Akaike information criterion during model optimization to minimize overfitting and (iv) provides mechanisms for comparing gene expression and cell fractions between samples in different classes. Computational microdissection of simulated and experimental tissue mixture datasets showed tight correlations between predicted and measured gene expression of pure tissues as well as tight correlations between reported and estimated cell fraction for each of the individual cell types. In simulation studies, MMAD showed superior ability to detect differentially expressed genes in mixed tissue samples when compared with standard metrics, including both significance analysis of microarrays and cell type-specific significance analysis of microarrays.

Conclusions: We have developed a new computational tool called MMAD, which is capable of performing robust tissue microdissection in silico, and which can improve the detection of differentially expressed genes. MMAD software as implemented in MATLAB is publically available for download at http://sourceforge.net/projects/mmad/.

PubMed Disclaimer

Figures

Fig. 1.
Fig. 1.
AICc correction during model fit improves estimates of gene expression. We simulated three datasets containing mixtures of colon adenocarcinoma cells, CD8+ T-cells and adipose cells in different mixing proportions. The average percentage of adipose cells was decreased in each simulation (10, 1, 0.1%, respectively). Without AICc correction, we note a marked decrease in the ability of MMAD to predict the gene expression profile of adipose cells when the fraction of adipose cells drops to 1% or below (A). AICc correction dramatically stabilizes predictions, even at cell fractions averaging 0.1 percent (B). (See also Supplementary Fig. S3.)
Fig. 2.
Fig. 2.
Computational microdissection of rat tissue mixture dataset (GSE19830). We compared the gene expression of pure liver, brain and lung tissue with estimates obtained by deconvoluting impure (mixed) tissue samples using MMAD and csSAM. Both MMAD and csSAM were constrained to use the investigator supplied values of f. We evaluated performance of MMAD without normalization adjustments (supervised approach) (A) and with normalization adjustments (semi-supervised approach) (B). Results were compared with csSAM deconvolution in both log-space (C) and linear space (D). We note that MMAD outperforms csSAM in both log-space and linear space; in particular, the residual variance is markedly reduced when MMAD is run in a semi-supervised manner using the bias-correction parameters
Fig. 3.
Fig. 3.
Estimation of cell fraction using MMAD in an immune mixture dataset (GSE11058). We estimated the fractional contribution of individual immune cells to mixed samples using gene expression profiles of pure immune cells (A). There is a slight scaling artifact, which can be filtered out by multiplying the gene expression profiles of pure cells by an appropriate constant (normalization artifact) (B). Results are similarly robust using predefined characteristic gene subsets (C) or with a blind deconvolution using the top 1% most variable genes (D)
Fig. 4.
Fig. 4.
Comparison of approaches for estimation of cell fraction. MMAD compares favorably with all comparators. We note comparable performance between MMAD and quadratic deconvolution (Gong et al., 2011) after rescaling (renormalizing) the reported gene expression of pure cell types (r > 0.99 for both approaches). Performance is also comparable for deconvolution with target gene subsets between MMAD and DSA (Zhong et al., 2013). MMAD outperforms deconf() (Repsilber et al., 2010) in blind deconvolution
Fig. 5.
Fig. 5.
Detection of differentially expressed genes is improved with MMAD. The global MMAD test statistic provides greater discriminatory power than the SAM test-statistic for the detection of differentially expressed genes in mixed tissue samples with highly variable and moderately variable cell type composition; differences are not seen when variability in cell type fraction is low (A). In a moderately variable simulation, MMAD and csSAM perform similarly for detecting differences in differential expression for the major cell type present in the simulated mixed tissue samples (colon adenocarcinoma cells, average cell fraction 0.7); MMAD shows improved ability to detect differential expression in cell types present at lower frequencies in the simulation (CD8+ T-cells and adipose cells, average cell fractions 0.2 and 0.1) (B)
See this image and copyright information in PMC

References

    1. Abbas AR, et al. Deconvolution of blood microarray data identifies cellular activation patterns in systemic lupus erythematosus. PLoS One. 2009;4:e6098. - PMC - PubMed
    1. Ahn J, et al. Demix: deconvolution for mixed cancer transcriptomes using raw measured data. Bioinformatics. 2013;29:1865–1871. - PMC - PubMed
    1. Alizadeh AA, et al. Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling. Nature. 2000;403:503–511. - PubMed
    1. Beer DG, et al. gene-expression profiles predict survival of patients with lung adenocarcinoma. Nat. Med. 2002;8:816–824. - PubMed
    1. Belbin TJ, et al. Molecular classification of head and neck squamous cell carcinoma using cDNA microarrays. Cancer Res. 2002;62:1184–1190. - PubMed

Publication types