JDSNMF: Joint Deep Semi-Non-Negative Matrix Factorization for Learning Integrative Representation of Molecular Signals in Alzheimer’s Disease

Abstract

High dimensional multi-omics data integration can enhance our understanding of the complex biological interactions in human diseases. However, most studies involving unsupervised integration of multi-omics data focus on linear integration methods. In this study, we propose a joint deep semi-non-negative matrix factorization (JDSNMF) model, which uses a hierarchical non-linear feature extraction approach that can capture shared latent features from the complex multi-omics data. The extracted latent features obtained from JDSNMF enabled a variety of downstream tasks, including prediction of disease and module analysis. The proposed model is applicable not only to sample-matched multiple data (e.g., multi-omics data from one cohort) but also to feature-matched multiple data (e.g., omics data from multiple cohorts), and therefore it can be flexibly applied to various cases. We demonstrate the capabilities of JDSNMF using sample-matched simulated data and feature-matched multi-omics data from Alzheimer’s disease cohorts, evaluating the feature extraction performance in the context of classification. In a test application, we identify AD- and age-related modules from the latent matrices using an explainable artificial intelligence and regression model. These results show that the JDSNMF model is effective in identifying latent features having a complex interplay of potential biological signatures.

Interpretable machine learning for dementia: A systematic review

“…Design the entire study with the end‐user in mind : The choice of interpretability method depends on the needs of the end‐user. Therefore, it can be beneficial to conceptualize the type of questions to be asked, whether that may be “which features are most important to the model” or “for this individual, how have the input features been used to arrive at the final prediction?” Addressing the interpretability of the study early on will allow researchers to better design their study, such as determining whether ground‐truth annotations may be desired to validate their interpretability models or if simulated preliminary results could benefit them as previously seen 56,58,66 . Research aiming to perform classification between disease groups may be better suited toward group‐level post hoc explanations that are able to highlight specific features of interest.…”

“…Therefore, it can be beneficial to conceptualize the type of questions to be asked, whether that may be “which features are most important to the model” or “for this individual, how have the input features been used to arrive at the final prediction?” Addressing the interpretability of the study early on will allow researchers to better design their study, such as determining whether ground‐truth annotations may be desired to validate their interpretability models or if simulated preliminary results could benefit them as previously seen. 56 , 58 , 66 Research aiming to perform classification between disease groups may be better suited toward group‐level post hoc explanations that are able to highlight specific features of interest. Alternatively, if the focus of the research is to better understand the disease‐causing pathology, then counterfactual examples that provide clinicians with an explanation of which brain changes would convert a diagnosis from healthy to dementia may be more useful.…”

Identification of the robust predictor for sepsis based on clustering analysis

“…Many methods, including K-means clustering, auto-encoder 30 , and NMF, have been used to construct groups, clusters, or modules. Moon and Lee developed the integrative non-linear representation method (Joint deep semi-NMF, JDSNMF) based on deep learning and modified NMF considering both intensity and direction of feature for the clustering task 54 . We implemented a widely used and simple method, K-means clustering to establish several subgroups.…”

“…We implemented a widely used and simple method, K-means clustering to establish several subgroups. Future study implementing integrative algorithm 30 , 54 for clustering is needed to establish the sophisticated module.…”

Predicting Alzheimer’s Disease with Multi-Omic Data: A Systematic Review

“…Types of models common to these studies are the SVM, RF, and XGBoost. Bespoke models published in these studies include DeepGAMI [31], JDSNMF [32], MSLPL [33] [34] was the only study to use mouse models to validate their conclusions in vivo.…”