Entropy-Guided Sample-Specific Feature Selection for Robust Incomplete Multi-Omics Learning in Gut Microbiome Disease...

Area of Science:

• Computational biology
• Bioinformatics
• Systems biology

Background:

• Multi-omics integration offers deep insights into complex diseases.
• Challenges include incomplete data, heterogeneity, and high dimensionality, hindering robust analysis.
• Accurate disease prediction and biomarker discovery are crucial for personalized medicine.

Purpose of the Study:

• To propose a novel framework, entropy-guided sample-specific feature selection for robust incomplete multi-omics learning (ESSFS-IMO), for accurate disease prediction and interpretable biomarker discovery.
• To address limitations of incomplete modalities, heterogeneity, and high dimensionality in multi-omics data.
• To develop a robust method for analyzing multi-omics data under missing-data conditions.

Main Methods:

• Combines instance-wise feature selection, entropy-adaptive optimization, and variational representation learning.
• Utilizes a Gumbel-Softmax-based selector for per-sample differentiable feature selection, guided by an entropy-based annealing strategy.
• Integrates selected features via an information-bottlenecked variational backbone with variance-weighted fusion for robust classification.

Main Results:

• ESSFS-IMO outperforms state-of-the-art baselines in accuracy, F1-score, and AUC on inflammatory bowel disease datasets.
• The model maintains high performance across various missing data patterns.
• Identifies biologically coherent biomarkers linking microbial, transcriptional, and metabolic profiles to immune regulation.

Conclusions:

• ESSFS-IMO provides a robust and interpretable solution for incomplete multi-omics learning.
• The framework achieves superior predictive power and resilience by integrating entropy-guided selection and variational information bottlenecks.
• ESSFS-IMO holds promise for broader biomedical applications in analyzing complex diseases and identifying therapeutic targets.