About the course

This course provides an introduction to causal inference and causal representation learning, offering both theoretical foundations and hands-on training. You will learn how to apply these methodologies to various biomedical data types, including clinical, genotype-phenotype, molecular, and multimodal data.

Causal inference and causal representation learning are emerging fields in AI and biomedical data science, enabling a shift from associational to cause-and-effect reasoning. These approaches have significant applications in biomedicine, such as evaluating treatment effectiveness, understanding causal mechanisms, identifying genetic risk factors, and uncovering causal relationships in complex molecular datasets. For instance, they can help answer questions like how effective a treatment is in preventing disease, whether its effects are direct or mediated by intermediate variables, and which genetic variants causally increase disease risk and can be targeted by drugs.

Gain practical experience using widely adopted tools and resources to apply causal techniques in real-world biomedical contexts. This EMBO Practical Course will also feature keynote talks that will provide insights into the role of causality in genomic, molecular, and computational biology, highlighting recent advancements and future directions in the field.

Who is this course for?

This course is for you if you are a PhD student, post-doctoral researcher, or research scientist currently working with clinical healthcare and/or molecular data. You will find the event of interest if you are keen to gain a better knowledge of causal inference approaches and how these can be leveraged to understand biological and biomedical problems. You should be relatively new to the application of causality in biomedicine, and be working as a computational biologist or bioinformatician, quantitative molecular biologist, statistical geneticist, AI/ML and biostatistics researcher, or similar.

You will be expected to have a working knowledge of the Linux operating system and the ability to use the command line. Basic use of R, Python, or Julia programming languages are required. 

While the course will make use of simple coding or streamlined approaches such as R Markdowns or Python notebooks, higher levels of competency will allow participants to focus on the scientific methodologies rather than the practical aspects of coding and how they can be applied in their own research.

We recommend these free tutorials:

• Basic introduction to the Unix environment 
• Introduction and exercises for Linux 
• Python tutorial
• R tutorial 
• Julia tutorial

Regardless of your current knowledge, we encourage you to use these to prepare for attending the EMBO Practical Course, if you are successfully selected. You may  also be sent materials prior to the course if the trainers believe this would be helpful to your prior understanding of the topics. These might include pre-recorded talks and required reading that will be essential to fully understand the course.

Learning outcomes

After the course, you should be able to:

• Assess the difference between causal and associational estimation 
• Explain the difference between randomised experiments vs observational studies in the context of public health and the human ecosystem
• Perform nonparametric estimation techniques on real-world health data to estimate causal effects on biomedical data
• Examine challenges of causal discovery for high-throughput molecular studies
• Assess whether causality can be extracted from data and, if so, apply the appropriate technique to extract causality in experimental and observational settings
• Apply causal representation learning to extract causal (latent) features from molecular/imaging data

Course content

During this course, you will learn about: 

• Causal inference and applications to clinical and genotype-phenotype data
  • Estimating causal effects: why correlations alone are misleading
  • Data in causality: randomised trials vs observational data
  • Estimation techniques for causal effects, e.g., targeted learning
  • Applications in simple clinical settings 
• Causal representation learning and application to multimodal molecular data
  • Extracting causal structure from observations: assumptions and challenges
  • Multimodal learning for causal (latent) feature discovery
  • Application of these methods to multimodal molecular data
  • Synergies of causal inference and representation learning in biomedicine

Trainers

Sjoerd Beentjes - The University of Edinburgh, Nima Hejazi - Department of Biostatistics Harvard T.H. Chan School of Public Health,Ava Khamseh - The University of Edinburgh, Pablo Rodríguez Mier, University of Heidelberg, Asantewaa Sarpong - University of Edinburgh, Caroline Uhler - MIT and Broad Institute, Yuelin Yao - MRC Mitochondrial Biology Unit University of Cambridge, Flaminia Zane - EMBL-EBI, Xinyi Zhang - Broad Institute and MIT.