ArtiPro

Personalised medicine aims to predict therapeutic response according to a personal profile that includes clinical, biological, and genetic data. This project focuses on depression. It aims to establish an artificial intelligence platform that brings together data from clinical research on the components of these profiles with the purpose of identifying predictors for response to depression treatment. The results will be combined into a single data platform that enables the use of large multimodal datasets to develop predictive models of symptoms and outcome data, thus enhancing the impact of these data. Artificial intelligence approaches will be investigated to identify novel biomarkers that can predict response to treatment. This will help to develop of a decision support system for personalised therapy while identifying the specific ethical and legal requirements that need to be fulfilled.

The project achieved substantial advances in pharmacogenetic modelling and personalised pharmacotherapy. In collaboration withthe German an the Norwegian partner, the team developed increasingly sophisticated statistical models addressing the combined influence of genetic variation, drug metabolism, and drug-drug interactions. This work resulted in publications on CYP2C19-based dose modelling and drug-drug-gene interactions. The work was subsequently extended to gene-gene-drug interactions and therapeutic drug-monitoring data. A bespoke statistical model developed during the project makes it possible to estimate the effects of multiple concomitant medications on CYP2C19 activity in real-world clinical data. Importantly, the estimated effects of interacting drugs were shown to correspond to their independently established affinity for the enzyme, providing empirical validation of the approach and creating new possibilities for studying polypharmacy and pharmacogenetic phenoconversion in clinical practice. A second major achievement concerned the identification and methodological validation of clinically relevant neuroimaging phenotypes. Early work investigated sources of physiological noise in functional MRI, resulting in a detailed characterization of the relationship between the fMRI global signal and signals originating from cranial bone. This methodological work provided knowledge directly applicable to improving the reliability of individual-difference analyses and was shared with consortium partners. Building on this methodological foundation, the project generated important findings concerning cognitive effort, sustained attention, and self-regulation. Functional imaging studies characterized dopaminergic and cholinergic substrates involved in cognitive effort and reward. This work led to the identification of a neuroimaging phenotype associated with individual differences in self-regulation, localized to a brain region associated with basal forebrain cholinergic nuclei. The finding is particularly noteworthy because the contribution of the human cholinergic system to stable behavioural traits and clinically relevant self-regulation remains poorly understood. The key result in this area directly related to depression and antidepressant treatment. Using the same phenotype, neural correlates of escitalopram treatment were identified in the ventral tegmental area and dorsal raphe. These findings provide evidence for the simultaneous involvement of dopaminergic and serotonergic systems in depression treatment and demonstrate the feasibility of using functional MRI to assess activity associated with the dorsal raphe. This represents an important advance towards imaging-based characterization of individual differences relevant to treatment response. A further innovative achievement was the introduction of artificial-intelligence and language-model methods for integrating heterogeneous psychopathology data. The team developed an AI-based approach for deriving a synthetic index of personality-disorder severity from commonly used personality scales. This approach has the potential to facilitate joint analyses of mental-health datasets in which conceptually related but non-identical assessment instruments have been used.