Structure-revealing data fusion model with applications in metabolomics

Abstract

In many disciplines, data from multiple sources are acquired and jointly analyzed for enhanced knowledge discovery. For instance, in metabolomics, different analytical techniques are used to measure biological fluids in order to identify the chemicals related to certain diseases. It is widely-known that, some of these analytical methods, e.g., LC-MS (Liquid Chromatography - Mass Spectrometry) and NMR (Nuclear Magnetic Resonance) spectroscopy, provide complementary data sets and their joint analysis may enable us to capture a larger proportion of the complete metabolome belonging to a specific biological system. Fusing data from multiple sources has proved useful in many fields including bioinformatics, signal processing and social network analysis. However, identification of common (shared) and individual (unshared) structures across multiple data sets remains a major challenge in data fusion studies. With a goal of addressing this challenge, we propose a novel unsupervised data fusion model. Our contributions are two-fold: (i) We formulate a data fusion model based on joint factorization of matrices and higher-order tensors, which can automatically reveal common and individual components. (ii) We demonstrate that the proposed approach provides promising results in joint analysis of metabolomics data sets consisting of fluorescence and NMR measurements of plasma samples in terms of separation of colorectal cancer patients from controls.

The promise of metabolic phenotyping in gastroenterology and hepatology

“…43,44 Data-fusion methods have also been developed to interrelate and interpret multiple data matrices, such as NMR, clinical chemistry or gene expression, to achieve a more holistic picture of the dynamic and interactive biological processes in the individual. [45][46][47] Statistical correlation methods applied to spectroscopic data can be used to find candidate biomarkers for a disease, to identify drugs and their metabolites 48,49 and to establish pathway connections between molecules. Modifications of the basic statistical spectroscopy focus on the identification of metabolites in fairly small subsets, which are otherwise hidden in large patient numbers, or on the detection of idiosyncratic responses to drugs (for example, subset optimization by referencing matching, STORM).…”

Structure-revealing data fusion

“…While NMR can capture all chemicals, one of the chemicals is invisible to LC-MS. We demonstrate the effectiveness of our model on this prototypical example using real data, where coupled data sets have both shared and unshared components. This is an extended version of our work [25] where, we have introduced our model briefly and discussed preliminary findings in cancer metabolomics. Here, we study the performance of the model in depth using both simulated and real data sets, where the underlying ground truth is known.…”

Joint Decompositions with Flexible Couplings

“…We show some results on the coupling of two tensor models where the coupled factors do not have the same size due to different sampling. This type of coupling cannot be dealt with the flexible models presented in [2] or [16]. Note also that this type of model appears naturally in multimodal data fusion since nothing guarantees that measurement devices of different nature will generate data sets with the same resolution.…”

“…It is then not surprising that multimodal data fusion, i.e. fusion of heterogeneous data, has become an important topic of research in these domains [21,2,10].…”

“…A more flexible model for the coupling of the models has been proposed in [2]. Instead of considering equality constraints for the entire factors of a tensor model, only a few components are constrained.…”