An AO-ADMM Approach to Constraining PARAFAC2 on All Modes

Roald, Marie; Schenker, Carla; Calhoun, Vince D.; Bro, Rasmus; Cohen, Jérémy E.; Acar, Evrim

doi:10.1137/21m1450033

Cited by 12 publications

(10 citation statements)

References 48 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This paper demonstrated that SIT provides a fast, accurate alternative to existing PARAFAC2 algorithms 10,[12][13][14]18 for untargeted modeling of hyphenated chromatography data. In contrast to classical PARAFAC2-ALS, 10 SIT allows constraints in the shifted mode, for example, non-negativity.…”

Section: Discussionmentioning

confidence: 97%

“…Alternative implementations have been developed recently that enable the use of constraints in all modes. [12][13][14][15] Some of the developed approaches show unexpected properties (e.g., more components are required to achieve a suitable PARAFAC2 solution). 15 Although some important improvements have been made to increase computational efficiency, 16 most algorithms for calculating the PARAFAC2 model 13,16,17 are still comparatively slow.…”

Section: Introductionmentioning

confidence: 99%

“…This is because the constraints must be imposed on the product of two matrices and not on a single‐factor matrix. Alternative implementations have been developed recently that enable the use of constraints in all modes 12–15 . Some of the developed approaches show unexpected properties (e.g., more components are required to achieve a suitable PARAFAC2 solution) 15 .…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Shift‐invariant tri‐linearity—A new model for resolving untargeted gas chromatography coupled mass spectrometry data

Schneide,

Bro,

Gallagher

2023

Journal of Chemometrics

Self Cite

View full text Add to dashboard Cite

Multi‐way data analysis is popular in chemometrics for the decomposition of, for example, spectroscopic or chromatographic higher‐order tensor datasets. Parallel factor analysis (PARAFAC) and its extension, PARAFAC2, are extensively employed methods in chemometrics. Applications of PARAFAC2 for untargeted data analysis of hyphenated gas chromatography coupled with mass spectrometric detection (GC‐MS) have proven to be very successful. This is attributable to the ability of PARAFAC2 to account for retention time shifts and shape changes in chromatographic elution profiles. Despite its usefulness, the most common implementations of PARAFAC2 are considered quite slow. Furthermore, it is difficult to apply constraints (e.g., non‐negativity) to the shifted mode in PARAFAC2 models. Both aspects are addressed by a new shift‐invariant tri‐linearity (SIT) algorithm proposed in this paper. It is shown on simulated and real GC‐MS data that the SIT algorithm is 20–60 times faster than the latest PARAFAC2‐alternating least squares (ALS) implementation and the PARAFAC2‐flexible coupling algorithm. Further, the SIT method allows the implementation of constraints in all modes. Trials on real‐world data indicate that the SIT algorithm compares well with alternatives. The new SIT method achieves better factor resolution than the benchmark in some cases and tends to need fewer latent variables to extract the same chemical information. Although SIT is not capable of modeling shape changes in elution profiles, trials on real‐world data indicate the great robustness of the method even in those cases.

show abstract

Section: Discussionmentioning

confidence: 97%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Shift‐invariant tri‐linearity—A new model for resolving untargeted gas chromatography coupled mass spectrometry data

Schneide,

Bro,

Gallagher

2023

Journal of Chemometrics

Self Cite

View full text Add to dashboard Cite

show abstract

“…Future work may extend EHR-based phenotyping by incorporating other modalities such as various omics data sets, and jointly analyze those data sets using CMTF-based approaches. Recent advances in CMTF methods (e.g., different loss functions for different data sets, various constraints (Schenker et al, 2021), PARAFAC2 models with flexible constraints in all modes (Roald et al, 2022)) may facilitate the progress in that direction.…”

Section: Possible Future Directionsmentioning

confidence: 99%

“…Here,

I \in {normalℝ}^{R \times R}

denotes the identity matrix,

Δ \in {normalℝ}^{R \times R}

is common for all

{boldB}_{k}

k = 1, \dots, K

. Other algorithmic approaches have also been studied when constraints are needed on the factor matrices (Cohen & Bro, 2018; Roald et al, 2022). Within the context of temporal phenotyping, PARAFAC2 has been extended to model binary data (Yin, Afshar et al, 2020).…”

Section: Background: Phenotyping Via Low‐rank Approximationsmentioning

confidence: 99%

UnsupervisedEHR‐based phenotyping via matrix and tensor decompositions

Becker

Smilde

Acar

2023

WIREs Data Min & Knowl

Self Cite

View full text Add to dashboard Cite

Computational phenotyping allows for unsupervised discovery of subgroups of patients as well as corresponding co‐occurring medical conditions from electronic health records (EHR). Typically, EHR data contains demographic information, diagnoses and laboratory results. Discovering (novel) phenotypes has the potential to be of prognostic and therapeutic value. Providing medical practitioners with transparent and interpretable results is an important requirement and an essential part for advancing precision medicine. Low‐rank data approximation methods such as matrix (e.g., nonnegative matrix factorization) and tensor decompositions (e.g., CANDECOMP/PARAFAC) have demonstrated that they can provide such transparent and interpretable insights. Recent developments have adapted low‐rank data approximation methods by incorporating different constraints and regularizations that facilitate interpretability further. In addition, they offer solutions for common challenges within EHR data such as high dimensionality, data sparsity and incompleteness. Especially extracting temporal phenotypes from longitudinal EHR has received much attention in recent years. In this paper, we provide a comprehensive review of low‐rank approximation‐based approaches for computational phenotyping. The existing literature is categorized into temporal versus static phenotyping approaches based on matrix versus tensor decompositions. Furthermore, we outline different approaches for the validation of phenotypes, that is, the assessment of clinical significance. This article is categorized under: Algorithmic Development > Structure Discovery Fundamental Concepts of Data and Knowledge > Explainable AI Technologies > Machine Learning

show abstract