Synopsis

Diseases such as cancer, cardiovascular disorders, and neurological conditions are complex and involve interactions across multiple biological layers (genomics, transcriptomics, proteomics, etc.). Integrating multi-omics data provides a more comprehensive understanding of disease mechanisms and can significantly improve predictive accuracy for disease classification and progression. We aim to develop a deep learning-based system to classify diseases and predict their progression by integrating multiple omics modalities. The model will use multi-modal deep learning architectures to process different omics datasets (e.g., gene expression, DNA methylation, proteomics) through separate branches, fusing them at various levels (early, late, or intermediate fusion). Attention mechanisms will be employed to identify the most informative omics layers, while autoencoders will reduce dimensionality and noise. Graph Neural Networks (GNNs) may also be used to represent multi-omics data as interaction graphs. This approach will enable accurate disease classification, subtype discovery, and survival analysis, contributing to personalized medicine and improved patient outcomes.

Relevance of the Topic

Multi-omics integration is essential for capturing the complexity of biological systems and understanding disease mechanisms at multiple levels. Traditional single-omics analyses often fail to capture the full picture, leading to incomplete insights. By leveraging deep learning to integrate multi-omics data, we can uncover novel biomarkers, discover disease subtypes, and predict clinical outcomes with higher accuracy. This has significant implications for personalized medicine, early diagnosis, and treatment optimization.

Future Research/Scope

• Survival Analysis: Extend the model to predict patient survival times or time-to-event outcomes using techniques like Cox proportional hazards models or deep survival networks.
• Subtype Discovery: Use clustering techniques to identify novel disease subtypes based on multi-omics data, enabling more targeted therapies.
• Explainability: Incorporate explainability techniques to highlight key omics features driving disease classification or progression predictions.
• Cross-Disease Applications: Apply the model to other diseases beyond cancer, such as neurodegenerative or autoimmune disorders, to test its generalizability.
• Real-Time Monitoring: Develop tools for real-time monitoring of disease progression using longitudinal multi-omics data.

Skills Learned

• Deep Learning: Hands-on experience with multi-modal neural networks, attention mechanisms, autoencoders, and GNNs for multi-omics data integration.
• Bioinformatics: Understanding of omics data preprocessing, normalization, and feature extraction.
• Python Programming: Proficiency in Python and libraries like TensorFlow, PyTorch, and Scikit-learn for implementing deep learning models.
• Data Visualization: Skills in visualizing multi-omics data and model predictions using tools like Matplotlib, Seaborn, and Cytoscape.

Relevant courses to the topic

• CSE422: Artificial Intelligence
• CSE425: Neural Networks
• CSE427: Machine Learning
• CSE443: Bioinformatics I
• Coursera: AI for Medical Diagnosis
• Coursera: Genomic Data Science Specialization

Reading List

• Books
  • "Deep Learning for Biomedical Data Analysis" – Mourad Elloumi (Link)
• Research Papers
  • "A review of multi-omics data integration through deep learning approaches for disease diagnosis, prognosis, and treatment" – Wekesa et al., Frontiers in Genetics
    Link
  • "Deep learning-based approaches for multi-omics data integration and analysis" – Ballard et al., BioData Mining
    Link
  • "Multi-Omics Integration For Disease Prediction Via Multi-Level Graph Attention Network And Adaptive Fusion" – Luo et al., IEEE Journal of Biomedical and Health Informatics
    Link | Code
  • "MODILM: towards better complex diseases classification using a novel multi-omics data integration learning model" – Zhong et al., BMC Medical Informatics and Decision Making
    Link
  • "Multi-omics integration method based on attention deep learning network for biomedical data classification" – Gong et al., Computer Methods and Programs in Biomedicine
    Link
• Datasets
  • TCGA (The Cancer Genome Atlas): A Resource for Multi-Omics Data
    Link
  • GTEx (Genotype-Tissue Expression): A Resource for Tissue-Specific Gene Expression
    Link
  • ICGC (International Cancer Genome Consortium): A Global Resource for Cancer Genomics
    Link
• Code Tutorials & Repositories
  • Integrative analysis of single-cell multi-omics data using deep learning (with video tutorials) (Towards Data Science)
    Link
  • How to build a machine learning model to predict antimicrobial peptides (End-to-end Bioinformatics) (YouTube/GitHub)
    Link
• Videos & Playlists
  • "Machine Learning in Computational Biology" – MIT
    YouTube Playlist
  • "The What and Why of Multi-Omics Integration | 2023 EMSL Summer School, Day 5" – YouTube
    Link