[Paper Review] Deep IDA: A Deep Learning Method for Integrative Discriminant Analysis of Multi-View Data with Feature Ranking -- An Application to COVID-19 severity

Jiuzhou Wang, Sandra E. Safo|PubMed|Nov 18, 2021

COVID-19 Clinical Research Studies31 references4 citations

TL;DR

Deep IDA is a deep learning framework that jointly models nonlinear relationships among multi-omics views (e.g., proteomics, transcriptomics, metabolomics) and discriminates COVID-19 severity groups (ICU vs. non-ICU). It uses deep neural networks to learn nonlinear projections that maximize view association and class separation, while incorporating a bootstrap-based feature ranking method to identify biologically interpretable molecular signatures, outperforming state-of-the-art methods in classification and revealing enriched pathways linked to infection, cancer, and metabolic disorders.

ABSTRACT

COVID-19 severity is due to complications from SARS-Cov-2 but the clinical course of the infection varies for individuals, emphasizing the need to better understand the disease at the molecular level. We use clinical and multiple molecular data (or views) obtained from patients with and without COVID-19 who were (or not) admitted to the intensive care unit to shed light on COVID-19 severity. Methods for jointly associating the views and separating the COVID-19 groups (i.e., one-step methods) have focused on linear relationships. The relationships between the views and COVID-19 patient groups, however, are too complex to be understood solely by linear methods. Existing nonlinear one-step methods cannot be used to identify signatures to aid in our understanding of the complexity of the disease. We propose Deep IDA (Integrative Discriminant Analysis) to address analytical challenges in our problem of interest. Deep IDA learns nonlinear projections of two or more views that maximally associate the views and separate the classes in each view, and permits feature ranking for interpretable findings. Our applications demonstrate that Deep IDA has competitive classification rates compared to other state-of-the-art methods and is able to identify molecular signatures that facilitate an understanding of COVID-19 severity.

Motivation & Objective

• To address the limitations of linear and unsupervised multi-view methods in modeling complex, nonlinear relationships between multi-omics data and COVID-19 severity.
• To develop a one-step deep learning method that simultaneously maximizes association among multiple views and separation between disease groups (ICU vs. non-ICU).
• To incorporate feature ranking into a nonlinear multi-view learning framework to enable biological interpretability of identified molecular signatures.
• To improve classification performance in high-dimensional, low-sample-size scenarios typical in multi-omics studies.

Proposed method

• Deep IDA employs deep neural networks to nonlinearly transform each omics view into lower-dimensional representations.
• It formulates an optimization problem that maximizes linear correlation between the transformed views and simultaneously maximizes linear separation within each view.
• A bootstrap-based resampling technique is used to rank features by their contribution to view association and class separation, enabling interpretable feature selection.
• The method is implemented in Python and includes a two-stage process: first, top-ranked features are selected via resampling; second, a new deep model is trained on these features to generate final low-dimensional representations.
• The framework supports binary or categorical outcomes and is extensible to other outcome types.
• The approach is validated through simulations and applied to real multi-omics data from COVID-19 patients, with results compared to state-of-the-art methods.

Experimental results

Research questions

• RQ1 Can a deep learning framework jointly model nonlinear associations among multiple omics views and classify disease severity more effectively than existing linear or unsupervised methods?
• RQ2 Which molecular features (proteins, metabolites, transcripts) contribute most to the association among views and separation of COVID-19 patient groups?
• RQ3 Can the proposed feature ranking method identify biologically meaningful signatures that explain the molecular architecture of COVID-19 severity?
• RQ4 How does Deep IDA perform in high-dimensional, low-sample-size settings common in multi-omics data?
• RQ5 What biological pathways are enriched among the top-ranked features, and do they align with known mechanisms of SARS-CoV-2 pathogenesis?

Key findings

• Deep IDA achieved competitive classification accuracy compared to state-of-the-art linear and nonlinear methods, even in high-dimensional, small-sample settings.
• The method successfully identified a set of top-ranked molecules—proteins, metabolites, and transcripts—that better discriminated between ICU and non-ICU COVID-19 patients.
• Enrichment analysis of top-ranked features revealed significant associations with pathways related to cancer, neurological disorders, infectious diseases, and metabolic diseases.
• The bootstrap-based feature ranking method provided stable and interpretable feature importance scores, enabling biological insight beyond model performance.
• The top-ranked features were consistent across bootstrap replicates and highlighted molecules involved in lipid transport and immune dysregulation, aligning with prior findings in severe COVID-19.
• The method demonstrated robustness in simulations and real data applications, with the potential to guide further experimental validation of key molecular drivers of disease severity.