Mapping the behavioural causes of weight change variability with genetic lottery

Introduction

Obesity is a heritable chronic condition, costing 2% of GDP worldwide. Many obesity treatments—behavioural, pharmacological, and surgical—have been developed. Intriguingly, people's responses to treatments vary widely; some may lose a lot of weight whereas others may even gain weight!

Weight Change Variability

To predict such weight change variability, close to 200 measures have been proposed in the past. The measures can be organized into a PEBBL framework across five domains:

• Psychosocial
• Environment
• Behavioural
• Biology
• Life quality

Still, there are too many measures to be used as predictors or intervention targets.

Proposed Solution

To move the field forward, we propose a novel 3-step OBECAUSE pipeline consisting of consolidation, genomic causation, and validation.

1. Consolidation

In consolidation, we will use machine learning to find the best-predicting PEBBL measures in several large-scale weight loss datasets. The PEBBL measures will be integrated into a new PEBBL short questionnaire with wide coverage and good psychometric properties. The questionnaire will then be distributed to all participants of the Estonian Biobank to study the genomics of PEBBL.

2. Genomic Causation

For genomic causation, we will detect genetic variants behind PEBBL measures and weight change. Knowing these variants enables discovering additions to the PEBBL framework through genetic correlations and functional mapping. Importantly, as genetic variants are randomized through genetic lottery, they enable systematic causal mapping of PEBBL measures that have causal effects on weight change.

3. Validation

For validation, these causal measures will be used as inputs to design an OBECAUSE toolbox of weight loss interventions. The value of these interventions will be tested in a commercial weight loss app.

Conclusion

In summary, the OBECAUSE pipeline of narrowing scattered associations down to potential causal mechanisms with machine learning and genomic causal inference will set a new standard for the behavioural health sciences, allowing for quicker discovery of intervention targets.