Causal Machine Learning to Discover Biochemical Determinants of Physical Fitness

Abstract

Identifying the key pathways relevant to cardiorespiratory fitness is of great importance for both predicting exercise responsiveness and potentially finding which interventions are likely to affect it. While contemporary deep learning models have demonstrated great success in pattern recognition and generation for various data modalities, their ability to decipher the causal mechanisms underlying these patterns is limited. This work proposes and evaluates a methodology using state-of-the-art causal discovery and causal inference methods to uncover the relationships between different proteins and their impact on changes in individuals’ maximal oxygen consumption (a proxy for physical fitness).