A neuro-computational assessment of mood fluctuations and their impact on decision making – Moodeling

Life events – whether positive or negative – induce mood variations, while being in a good mood would induce a “rosy outlook”, leading us to view events (or prospects) as more positive than they objectively are. Such reciprocal interaction is also supported by clinical observations. We and others recently used computational modeling to study this phenomenon: we showed that mood can be described as a leaky accumulation of prediction errors about feedbacks. The influence of feedback on mood is reciprocal, the same feedback being more positively perceived when mood is higher. Different individuals can integrate the same sequence of events in different ways, resulting in different mood levels. We coupled this approach with functional MRI and demonstrated that mood was reflected in the baseline activity of two regions: ventromedial prefrontal cortex (vmPFC) and bilateral anterior insula (aIns), with positive and negative encoding respectively. However, in this preliminary works, the time scale of induced mood fluctuations was around a couple of minutes, which is very short compared to everyday mood fluctuations. Their amplitude was also extremely small compared to pathological (or even normal) ones.

The goal of this project is to bridge the gap between lab experiments and everyday normal but also pathological mood fluctuations by operating a triple change of scale: in terms of number of participants, time-scale, and amplitude. More specifically, we aim to test (1) to what extent the very same computational principle could underlie mood fluctuations at these very different time scales (2) to what extent our neuro-computational approach of mood fluctuations and its impact on decision-making can discriminate patients with different types of mood disorder from healthy participants, and predict their clinical outcome and (3) to what extent vmPFC and aIns baseline activities reflect not only short-term minimal experimentally induced mood fluctuations but also every day normal mood fluctuations, as well as day-to-day variations in decision-making.

In a first work package, we will analyze three large datasets: subjective mood ratings collected in the general population, a validated depression questionnaire completed by more than 400 000 participants over the past 17 years, and clinically relevant variables (such as the daily number of consultations for depression, or the daily number of suicide attempts) extracted from a clinical data warehouse. We aim to demonstrate that the same computational model accounts both for short-time mood fluctuations in the context of a lab experiment and long-term normal and pathological mood fluctuations at a population scale

In a second work package, we will compare two groups of patients with mood disorders, bipolar disorder or recurrent depressive disorder, to healthy controls using a neuro-computational approach. Critically, we will combine a short-term evaluation coupled with fMRI and a long-term follow-up thanks to a dedicated smartphone application. We aim to demonstrate that mood disorders are characterized by a specific computational fingerprint describing how of positive and negative events are integrated into a mood signal, which in return affects decision-making. Moreover, we will use this computational fingerprint to predict clinical outcomes.

Finally, we will rely on a very specific condition, patients with drug-resistant epilepsy for which stereotactical EEG is required, to obtain a continuous recording of our two regions of interest over a few days. This will allow us to investigate the brain correlates of mood and their impact on decision-making at a relevant time-scale.