Analyzing multi-way measurements with variations across one mode of the dataset is a challenge in various fields including data mining, neuroscience and chemometrics. For example, measurements may evolve over time or have unaligned time profiles. The PARAFAC2 model has been successfully used to analyze such data by allowing the underlying factor matrices in one mode (i.e., the evolving mode) to change across slices. The traditional approach to fit a PARAFAC2 model is to use an alternating least squares-based algorithm, which handles the constant cross-product constraint of the PARAFAC2 model by implicitly estimating the evolving factor matrices. This approach makes imposing regularization on these factor matrices challenging. There is currently no algorithm to flexibly impose such regularization with general penalty functions and hard constraints. In order to address this challenge and to avoid the implicit estimation, in this paper, we propose an algorithm for fitting PARAFAC2 based on alternating optimization with the alternating direction method of multipliers (AO-ADMM). With numerical experiments on simulated data, we show that the proposed PARAFAC2 AO-ADMM approach allows for flexible constraints, recovers the underlying patterns accurately, and is computationally efficient compared to the state-of-the-art. We also apply our model to a real-world chromatography dataset, and show that constraining the evolving mode improves the interpretability of the extracted patterns.

Analysis of time-resolved postprandial metabolomics data can enhance our knowledge about the human metabolism by providing a better understanding of regulation of subgroups of metabolites (e.g., lipids) and variations in postprandial responses of subgroups of people, with the potential to ultimately advance precision medicine. However, characterizing postprandial metabolomics response and understanding group differences is a challenging task since it requires the analysis of large-scale metabolomics data from a large set of individuals containing measurements of a wide set of metabolites at multiple time points. Such data is in the form of a three-way array: subjects by metabolites by time points. The state-of-the-art analysis methods mainly focus on clustering temporal profiles relying on summaries of the data across subjects or univariate analysis techniques studying one metabolite at a time, and fail to associate subgroups of subjects and subsets of metabolites with the dynamic time profile simultaneously. 

In this study, we use NMR (nuclear magnetic resonance) spectroscopy measurements of plasma samples (of over three hundred individuals from the COPSAC2000 cohort) collected at multiple time points during a challenge test.  We use a multi-way analysis technique called the CANDECOMP/PARAFAC (CP) model to extract interpretable patterns from the time-resolved data. We compare the analysis of postprandial data, fasting state-corrected data and only fasting state data, and demonstrate the differences between different analysis approaches.

Our results show that the CP model reveals biologically meaningful patterns capturing how certain metabolite groups and their temporal profiles relate to various meta variables, in particular, BMI (body mass index), confirming already known biological knowledge as well as revealing new biological insights.

An effective way of jointly analyzing data from multiple sources is through coupled matrix and tensor factorizations (CMTF). Different characteristics of datasets from multiple sources require to employ various regularizations, constraints, loss functions and different types of coupling structures between datasets. While existing algorithmic approaches for CMTF can incorporate constraints, linear couplings and different loss functions, none of them has been shown to achieve the flexibility to incorporate all. We propose a flexible algorithmic framework for coupled matrix and tensor factorizations, which utilizes Alternating Optimization (AO) and the Alternating Direction Method of Multipliers (ADMM). The framework facilitates the use of a variety of constraints, loss functions and couplings with linear transformations. Numerical experiments on simulated datasets and real data from chemometrics and hyperspectral super-resolution demonstrate that the proposed approach is accurate, flexible and computationally efficient with comparable or better performance than available CMTF algorithms.

While we focus on CANDECOMP/PARAFAC (CP) –based CMTF models, we will also briefly discuss the use of an AO-ADMM based algorithmic approach for fitting a PARAFAC2 model.  We demonstrate that the proposed algorithmic approach enables imposing constraints in all modes, which has been a challenge using the traditional alternating least squares-based algorithm used for PARAFAC2.