HINN: Hierarchical Input Neural Network identifies multi-omics biomarker for cognitive decline

Area of Science:

• Computational biology
• Genomics
• Neuroscience

Background:

• Complex diseases require models integrating multi-omics data for biological interpretability.
• Existing machine learning (ML) approaches often overlook hierarchical and regulatory relationships between molecular layers.

Purpose of the Study:

• To present the Hierarchical Input Neural Network (HINN), a deep learning framework incorporating cross-omics relationships.
• To improve predictive performance and biological interpretability in multi-omics data analysis.
• To identify blood-based biomarkers for Alzheimer's disease and cognitive decline.

Main Methods:

• Developed HINN, a deep learning framework embedding known cross-omics relationships (genomics, epigenomics, transcriptomics).
• Applied HINN to blood-derived multi-omics data from individuals with Alzheimer's disease or mild cognitive impairment.
• Predicted cognitive scores using standardized assessments and compared HINN against baseline and state-of-the-art models.

Main Results:

• HINN demonstrated superior performance in predicting cognitive scores compared to existing models.
• Identified multi-omics biomarkers, including SNPs and CpG sites in ATP6V1C1 and RCHY1, correlated with plasma p-Tau181 levels.
• Discovered biologically relevant features linked to cognitive decline processes.

Conclusions:

• HINN effectively integrates multi-omics data and biological knowledge for enhanced predictive performance and interpretability.
• The study highlights the potential of HINN in uncovering interpretable, blood-based biomarkers for complex diseases like Alzheimer's.
• Findings suggest HINN can advance the understanding and early detection of cognitive decline.