Revolutionizing Healthcare: Core Concept Session on Multiomics

Introduction

• Core Concept Session: Focus on fundamental ideas and methods from machine learning and medicine that transcend specific specialties.
• Topic: Multiomics - using high throughput techniques to understand the human body comprehensively.
• Relevance: Multi-dimensional data landscape suitable for machine learning.
• Hosts: Professor Mianda (University of Cambridge) & Tim Olink (Medical Student, KU Leuven).

Panelists Introduction

• Professor Andreas Floto: Respiratory Biology, University of Cambridge, research on immune responses to bacterial infections.
• Professor Kunu: Biomedical Informatics at Harvard Medical School, expertise in medical sciences and technology integration.
• Professor Julio S. Rodriguez: Medical Bioinformatics, Heidelberg University, focuses on computational biomedicine.
• Dr. Fergus Emy: Postdoc at UCLA, expertise in computational methods.

Agenda

1. Introduction to Multiomics
2. Panelist Presentations
3. Presentation of Work at the Fare Lab
4. Q&A Session
5. Closing Remarks

What is Multiomics?

• Information Flow in Biology: DNA → RNA → Proteins → Cellular Processes.
• Multiomics: Examining genomics, transcriptomics, proteomics, metabolomics, etc., to understand disease mechanisms.
• Applications:
  • Causal Discovery
  • Mechanisms of Disease
  • Biomarker Discovery
  • Synthetic Biology
  • Drug Discovery

Panelist Presentations

Professor Julio S. Rodriguez

• Focus on using prior knowledge to derive mechanisms from multiomic data.
• Importance of public data and databases to enhance analysis.
• Example: Study on myocardial infarction using single cell and spatial data.
• Future: Leveraging spatial multiomics data for better disease understanding.

Professor Kunu

• Developed AI for pathology and histopathology integration with multiomic data.
• Example: Real-time pathology diagnosis for brain cancer.
• Challenges: Implementing AI in clinical settings; requires collaboration.

Dr. Fergus Emy

• Computational challenges in multiomics: Dimensionality, limited label data, inter-feature correlation.
• Methods:
  • Vime for self-supervised learning.
  • Sefs for feature selection.
  • Deep IMV for integrating multiomics with missing data.
  • GCIT for conditional independence testing.

Q&A Highlights

• Challenges of Multidimensionality: Effective data integration and analysis remain complex.
• Application in Rare Diseases: Difficulties due to lack of data; AI methods to flag anomalies.
• Interpretability: Critical for clinical application; requires models to be understandable to clinicians.
• Future Directions:
  • More causal, robust, and fair machine learning methods.
  • Batch effect correction to separate biological information from experimental noise.

Conclusion

• Next Sessions: Intensive Care Medicine Spotlight & Core Concept Session on Treatment Effect Estimation.
• AI in Medicine Summer School: Targeted at clinicians and medical students, covering fundamentals of machine learning in healthcare.

Note: The session highlighted the potential and challenges of integrating multiomics with AI for transformative healthcare applications.