Computational modelling and assessment of depression: from neural mechanisms and etiology to measurable behaviour

Depression is a highly prevalent clinical condition which has been estimated to affect a growing part of the population in western countries. Alongside expenditure on diagnostics and treatment, there is a high economic impact due to lost productivity. Although a range of treatments are available, diagnoses are currently costly and require subjective assessment by a specialist. Moreover, treatment selection can be lengthy and can involve trial and error. To develop better diagnostics, stratification, and treatments for depression, we need a better understanding of the condition across different levels – from neural mechanisms to cognition and behaviour. Computational modelling is an emergent theory-driven approach which can aid linking data across different levels of analysis – from neural mechanisms and computations in the brain, to cognitive algorithms and observable behaviour. Some models integrate diverse findings and make predictions, while others enable inference of clinical measures which are not obvious in raw data. Modelling can lead to better understanding of depression, and in turn to better stratification and treatments. On the other hand, machine learning and classification methods can help detect clinically-relevant patterns in experimental data in a purely data-driven manner. This can lead to development of better screening and diagnostic methods. In the current work, we first review some of the most prominent neurocognitive theories of depression, as well as existing studies which used computational modelling methods. Based on our review, we argue that modelling can provide a rich set of tools for a better understanding of the condition. We then develop two novel computational modelling accounts of depression. In the first account, we propose an explicit mechanistic link between a robust behavioural negative bias effect and some of the widely reported or theorised neural aspects of depression – hyperactive amygdala and inhibited dopamine release. In the second account, we attempt to better explain depressive cognitive deficits and show how they can arise from depression-relevant etiological factors – altered valuation and controllability estimates. Finally, in the third part of this work we attempt to develop a novel system for detecting depressive symptoms based on a combination of face-tracking, eye-tracking and cognitive performance measures. We evaluate the system in a pilot experiment and show that a combination of measures can achieve better results than measures from each domain separately