11-08, 14:35–15:15 (US/Eastern), Music Box
Alzheimer’s disease (AD), the predominant form of dementia worldwide, is a progressive neurodegenerative disease characterized by severe cognitive decline. Despite the widespread prevalence of AD, and significant investment into AD research, our understanding of this disease is still evolving owing to the complex interplay of the genetic and molecular mechanisms involved. Effective therapeutic development for AD requires a deeper understanding of these underlying interactions. In this talk, we first demonstrate how graph neural networks are utilized to integrate multi-modal molecular data (i.e. multi-omics data) to build predictive models of AD. We then apply machine learning interpretability methods to our models to generate insights into the disease’s underlying drivers and reveal key biological interactions. We discuss the technical challenges encountered, the limitations of our current approach, and potential future directions for advancing research in this field.
This talk is developed as a brief introduction to the use of AI in biomedical research for a general audience. In particular, we discuss our work on using GNNs to integrate and analyze multi-modal molecular data and derive disease-relevant insights. The goal of this talk to give the audience a broad introductory overview of molecular biology, Alzheimer's disease, multi-modal molecular data, and the technical challenges associated with the analysis of such data and how advanced AI methods such as GNNs can help.
Here is an outline for this talk and an approximate breakdown of time:
1. Speaker introduction and background - 2 minutes.
2. Introduction to Alzheimer's disease - brief overview of the disease (symptoms, hallmarks, relevance), why is AD research hard? - 2 minutes.
3. A brief introduction to computational biomedicine - basic principles of molecular biology and its relevance to the study of diseases such as AD - 2 minutes.
4. Multi-omics data - where does our data come from, how is data from multiple molecular layers collected and curated - 3 minutes.
5. Integrating multi-omics data into predictive models - why is it important? what are the computational challenges? - 4 minutes.
6. GNNs - what is a GNN? why is it useful here? - 4 minutes.
7. How do we use GNNs integrate them data from different molecular modalities? - 4 minutes.
8. Interpretability - how do we derive insights into the disease? what are the implications for therapeutic development - 4 minutes.
9. Limitations and directions for future work - 2 minutes.
10. Q&A- 3 minutes.
No previous knowledge expected
- Wikipedia entry for multi-omics data
- PyTorch Geometric - pytorch based library for building and training GNNs
- Alzheimer's Association explainer on Alzheimer's disease
- Bioarxiv preprint of paper describing multi-omics integration to derive insights in Alzheimer's using GNNs
- Review article on the integration and analysis of multi-omics data
- NIH-NIA explainer on Alzheimer's disease
