Deblender: a semi−/unsupervised multi-operational computational method for complete deconvolution of expression data from heterogeneous samples

Table 1 Summary of the features of the proposed Deblender and the methods examined for comparison. With the term ‘input’ we refer to any type of information other than the mixed expression data

	Classification: Partial or Complete	Mathematical/ Statistical method	Input: reference/ signature expression data	Input: marker gene lists	No input	Output: Cell/tissue type-specific proportions (A)	Output: Cell/tissue type-specific expression profiles (S)	Output: number of cell/tissue types
MMAD [7]	C/P	Maximum likelihood/ Conjugate gradient	Yes	Yes	Yes^a	Yes	Yes	No
DSA [9]	C	Quadratic programming	No	Yes	No	Yes	Yes	No
NMF-CELLMIX [12]	C	Non-negative Matrix Factorization	No	Yes	No	Yes	Yes	No
CIBERSORT [16]	P	v-Support Vector Regression	Yes	No	No	Yes	No	No
DeconRNASeq [24]	P	Quadratic programming	Yes	No	No	Yes	No	Yes
Deblender	C	Least Squares/ Quadratic programming/ Non-negative Matrix Factorization/ Unified Particle Swarm Optimization	No	Yes	Yes	Yes	Yes	Yes

ISSN: 1471-2105