The promise of a model-based psychiatry: building computational models of mental ill health

Abstract

Computational models have great potential to revolutionise psychiatry research and clinical practice. These models are now used across multiple subfields, including computational psychiatry and precision psychiatry. Their goals vary from understanding mechanisms underlying disorders to deriving reliable classification and personalised predictions. Rapid growth of new tools and data sources (eg, digital data, gamification, and social media) requires an understanding of the constraints and advantages of different modelling approaches in psychiatry. In this Series paper, we take a critical look at the range of computational models that are used in psychiatry and evaluate their advantages and disadvantages for different purposes and data sources. We describe mechanism-driven and mechanism-agnostic computational models and discuss how interpretability of models is crucial for clinical translation. Based on these evaluations, we provide recommendations on how to build computational models that are clinically useful.

Figure 1:. Trade-offs between models and data sources

(A) Models differ in their transparency of the mechanisms, which determines their best use. Although most complex models often achieve higher predictive performance, white box models allow an understanding of the underlying mechanisms. (B) The choice of data source matters. High quality data (such as laboratory experimental studies) are often expensive (eg, functional MRI). Passive data collection is inexpensive, but the features are often unclear and not well defined. By transforming laboratory-based methods (eg, using gamification), substantially larger datasets can be collected at lower costs.

Figure 2:. Mechanistic models of brain function

Schematic representation of different levels of abstraction used in modelling brain functioning from spiking network models (A) to neural populations (B) to models incorporating multiple brain regions (C).

Figure 3:. Computational modelling of indecisiveness

(A) Laboratory information gathering task in which a participant is asked to determine which of the two colours is the more plentiful by drawing cards on the board. (B) This task-based measure of indecisiveness is linked to indecisiveness as assessed using traditional clinical interviews and showing ecological validity. (C) Computational modelling of drawing behaviour revealed that humans are suboptimal when making their decision, gathering too little information when it was cost-free, but gathering too much when information collection was costly. (D) Best fitting models showed that participants accumulate subjective costs that promote early decisions, and a bias in this accumulation process was driving the difference between participants with and without with obsessive compulsive disorder. (E) Gamification of this task allows the assessment of indecisiveness outside the laboratory in large samples of diverse backgrounds using smartphone apps, such as Brain Explorer. Parts B and D were reproduced from Hauser et al and were published under a creative commons attribution (CC BY). (E) from Brain Explorer app (www.brainexplorer.net). OCD=obsessive compulsive disorder.

Figure 4:. Bringing data sources together to improve modelling in psychiatry

Although most research has focused on single data sources for their models, bringing complementary data sources together can help improve model performance. Therefore, mechanism-driven model indicators can help with the interpretability of black box models. Substituting complex in-laboratory data sources with more readily available proxies, such as smartphone-based games, can help bring research-led findings into a real-world setting. These extended strategies might help build clinically useful models.