Independent Component Analysis for Unraveling the Complexity of Cancer Omics Datasets

Abstract

Independent component analysis (ICA) is a matrix factorization approach where the signals captured by each individual matrix factors are optimized to become as mutually independent as possible. Initially suggested for solving source blind separation problems in various fields, ICA was shown to be successful in analyzing functional magnetic resonance imaging (fMRI) and other types of biomedical data. In the last twenty years, ICA became a part of the standard machine learning toolbox, together with other matrix factorization methods such as principal component analysis (PCA) and non-negative matrix factorization (NMF). Here, we review a number of recent works where ICA was shown to be a useful tool for unraveling the complexity of cancer biology from the analysis of different types of omics data, mainly collected for tumoral samples. Such works highlight the use of ICA in dimensionality reduction, deconvolution, data pre-processing, meta-analysis, and others applied to different data types (transcriptome, methylome, proteome, single-cell data). We particularly focus on the technical aspects of ICA application in omics studies such as using different protocols, determining the optimal number of components, assessing and improving reproducibility of the ICA results, and comparison with other popular matrix factorization techniques. We discuss the emerging ICA applications to the integrative analysis of multi-level omics datasets and introduce a conceptual view on ICA as a tool for defining functional subsystems of a complex biological system and their interactions under various conditions. Our review is accompanied by a Jupyter notebook which illustrates the discussed concepts and provides a practical tool for applying ICA to the analysis of cancer omics datasets.

Establishment of a pancreatic adenocarcinoma molecular gradient (PAMG) that predicts the clinical outcome of pancreatic cancer

“…Histological and molecular classifications of PDX suggest PDAC diversity may be better represented by a continuum of differentiation that is as also followed at the molecular level. To establish a robust continuous molecular description of PDAC, we applied an unsupervised approach termed independent component analysis (ICA) previously shown to derive highly reproducible signatures from transcriptome profiles by extracting biologically relevant components [10,11]. Figure S2 illustrates the procedure used to uncover an RNA signature which, in essence, builds on the blind deconvolution of the PDX transcriptomic profiles to generate component spaces.…”

Factorial study of the RNA-seq computational workflow identifies biases as technical gene signatures

“…The heatmap should be read in a column-wise manner, looking at what information each expression mode is providing. The majority of the expression modes seem to be driven by biological information, which are the modes that are usually studied when using ICA with RNA-seq data 20,41 and the modes of interest for researchers using RNA-seq to gain insight into biology. To confirm the informational value of the biological modes, we illustrated the distribution of the pipelines along four selected biological modes, one for each tissue, in Figure 2C.…”

“…PCA considers the data to follow a multivariate Gaussian distribution whereas ICA seeks a linear combination of non-Gaussian distributions. ICA has previously been applied to RNA-seq quantification results to infer groups of genes displaying a shared behaviour across several datasets 19,20 . ICA is a long-sought answer to the cocktail party problem, where an unknown number of persons talk in a room where a known number of microphones are placed 21 .…”

Reference-free deconvolution of complex DNA methylation data – a systematic protocol

“…Confounding factors, such as age, sex or donor genotype, can have a strong influence on the methylome, and investigators might want to adjust for those in their analyses 24,25 . Therefore, we argue that accounting for confounders, using methods such as Independent Component Analysis (ICA) 26 , is crucial to obtain biologically relevant results. As the final data preparation step, a CpG subset selection determines sites that are linked to, for instance, cell type identity or any other phenotypic trait of interest.…”

“…DNA methylomes can be affected by various sources of variability, both of biological and technical nature that might mask the signals of interest. Independent component analysis (ICA) 26 is a data-driven dimensionality reduction method that performs a matrix decomposition, dividing the experimentally observed data matrix D mn into k independent signals S mk mixed with the coefficients of M kn :…”