Assessing the validity and reliability of computational phenotyping of mood

Mood can be understood as an affective state resulting from the integration, over time, of positive and negative outcomes. To capture intra- and inter-individual variability in mood fluctuations, computational models have been increasingly applied to self-reported mood ratings obtained during behavioural tasks. Such computational models of mood may be useful tools for understanding mood disorders. However, to be used in a clinical setting, their validity and reliability should be assessed. Recent versions of these models incorporate reciprocal interactions between mood and event perception, governed by specific parameters. Hence, it is critical to determine the extent to which estimated parameters depend on potentially arbitrary aspects of experimental design (e.g., feedback sequences) and to assess their psychometric test-retest stability. We used two widely established mood-induction tasks—a lottery task and a general-knowledge quiz—alongside a newly developed task, designed to allow precise experimental control over outcome sequences, while preserving participants’ perception that outcomes depended on their actions. Extensive numerical simulations were conducted to test the robustness of the computational models. To evaluate test-retest reliability, 163 healthy volunteers completed the tasks twice, separated by a two-week interval. Simulations demonstrated robust parameter recovery overall, though estimating the effect of mood on feedback perception proved more challenging. All tasks successfully induced mood fluctuations, accurately described by models employing leaky integration of feedback. Test-retest reliability was satisfactory for two important parameters, baseline mood and accumulated feedback weight, with significant correlations observed across most parameters. Furthermore, our newly developed task confirmed that mood-related parameter estimates remained largely unaffected by specific feedback sequences. Computational models of mood dynamics show robust validity and satisfactory test-retest reliability. Stable parameters, such as baseline mood and feedback weighting, endorse the application of these models in longitudinal studies, offering a reliable methodological basis for clinical research on mood disorders.Author summary: Our mood changes from moment to moment as we experience positive and negative events. Researchers increasingly use mathematical models to describe these changes and to understand why mood fluctuates differently across people. Such models could eventually help study mood disorders, but they first need to be tested carefully: do they capture real mood changes, and do they give stable results when the same person is tested again? In this study, healthy volunteers completed mood-inducing tasks twice, two weeks apart. Two tasks were already commonly used in this field, and we also developed a new task that allowed us to control the sequence of positive and negative feedback while maintaining participants’ impression that outcomes depended on their actions. We also ran numerical simulations to check whether the models could reliably recover the quantities they were designed to estimate. Overall, the models captured mood fluctuations well. Some quantities, especially baseline mood and the impact of recent feedback on mood, were reasonably stable over time. Our new task also showed that these estimates were not strongly influenced by the exact feedback sequence. These findings support the use of such models in future studies of mood and mood disorders.