Graph designs for deep learning-based multi-omics integration
Muhtasim Noor Alif1, Khandakar Tanvir Ahmed1, Sudipto Baul1
1Department of Computer Science, University of Central Florida, 4000 Central Florida Blvd., Orlando, FL 32816, United States.
Briefings in Bioinformatics
|July 31, 2026
Summary
Integrating multi-omics data is challenging. Graph-based deep learning offers a flexible approach to model complex biological interactions, paving the way for new multi-omics studies.
Area of Science:
- Computational Biology
- Bioinformatics
- Genomics
Background:
- Modern sequencing enables multi-omic data generation from single biological systems.
- Integrating diverse omics data into a unified model presents significant challenges.
- Graph-based deep learning provides a flexible framework for representing complex molecular interactions and sample relationships.
Purpose of the Study:
- To review graph construction and usage in multi-omics deep learning models.
- To organize methods based on graph properties and integration strategies.
- To summarize design choices for interpretability, data needs, and robustness.
Main Methods:
- Surveying graph construction and deep learning applications in multi-omics.
- Organizing methods by node schema, edge semantics, interaction type, integration strategy, graph context, and model architecture.
- Analyzing methods across bulk, single-cell, and spatial omics data.
Main Results:
- Detailed overview of graph-based deep learning strategies for multi-omics integration.
- Summary of strengths and weaknesses of various design choices.
- Evaluation of suitability for prediction and mechanism-oriented discovery.
Conclusions:
- Graph-based deep learning is a powerful strategy for multi-omics data integration.
- Understanding design choices is crucial for effective model development.
- A practical pipeline is proposed for constructing and evaluating multi-omics graphs.
