Leveraging machine learning to uncover the hidden links between trusting behavior and biological markers

Understanding the decision-making mechanisms underlying trust is essential, particularly for individuals with mental disorders who often experience difficulties in forming interpersonal trust. In this study, we aimed to explore biomarkers associated with trust-based decision-making through quantitative analysis. However, quantifying internal decision-making processes is challenging, as they are not directly observable. To address this, we developed a machine learning method based on a Bayesian hierarchical model to quantitatively infer latent decision-making parameters from behavioural data collected during a trust game. Applying this method to data from patients with major depressive disorder (MDD) and healthy controls (HCs), we estimated individualised model parameters that regulate trust-related decisions. The model successfully predicted participants' behaviours in the task. Although no significant group-level differences were observed in the estimated parameters between the MDD and HC groups, we uncovered hidden links between trust-related decision-making processes and specific blood biomarkers. Notably, metabolites such as 5-aminolevulinic acid, acetylcarnitine, and 2-aminobutyric acid were significantly associated with individual differences in trusting behaviour. These findings provide valuable insight into the biological basis of trust-based decision-making. They also offer a novel framework for integrating behavioural modelling with biomarker discovery, potentially informing the development of targeted interventions to enhance social functioning and overall well-being.