Editorial

Psychiatry today aims to understand the mechanisms underlying mental disorders by studying endophenotypes that are related to mental functions. New analytical technologies as well as new tools for data analysis enable us to advance in the field of neuropsychiatry. The complexity of neurobiology is reflected by the huge number of discrete observations possible through novel high-resolution experimental technologies such as imaging, neurophysiology, gene expression, and next-generation sequencing. The integration and interpretation of these independent technologies asks for the application of sophisticated mathematical and statistical methods, in particular required for the generation of structured systems biology models. For instance, the lack of connectivity of neuro-networks as a possible cause of pathologies in the brain of schizophrenic patients or persons at risk is a useful concept that can translate observed data into predictive models. On the one hand, technologies are available in order to identify deficits of connectivity. On the other, major methodological obstacles need to be addressed. The key challenge is to correlate any observed data to its pathophysiological substrate. The considerable intra-individual variability between subjects could blur diagnostically relevant information. Low signal-to-noise ratio of single parameters can be substantially increased by using multivariate discrimination analysis in order to extract high-dimensional patterns across biological samples. Support vector machines (SVM) are well established statistical methods. SVM have been successfully employed as biomedical diagnostic tools especially in the field of prediction of psychosis. Their primary strength is to provide optimized classifying for single individuals close to the cutting plane, rather than describing statistical group differences. Altogether, it is obvious that obtainable measurements cannot capture all the essential processes over the scales from the genetics of neurons, the cellular communication up to the brain as an organ. The observable state space of a system of 1011 cells is unlimited, thus, even with methods covering the spectrum from molecular details to disease phenotypes, an exhaustive exploration remains impossible. In order to understand the brain as the most complex system, coarse models are necessary to represent comprehensive abstract “reconstructions” to interpolate and extrapolate the observed data. Large-scale models remain vague with respect to their resolution of mechanistic details, and detailed models can fail to elucidate complex behavior. However, as examples from meteorology and ecology have shown, large-scale computational models can improve the accuracy of predictions even in situations of weak causality and stochastic variation. Modelling complex systems aims for the reduction of complexity, but established strategies to cope with incomplete, erroneous, and not-well-understood information in biology are not yet available. The epistemological foundations of modelling approaches need to be explored and many basic issues separating correlation from causation are only beginning to be recognised as relevant problems in disease research.

In psychiatry, the epistemic benefit of structured systems approaches has not yet replaced the “soft-science” tradition of often speculative and anecdotal reasoning. In order to overcome the present scepticism, we need to construct conceptual models on the basis of computational models and to develop computational models on the basis of conceptual models. Here, we present some approaches to tackle the problem to understand at least some symptoms of mental disorders such as schizophrenia as complex dynamic systems that can be systematically understood. These papers are based on a symposium on systems biology and schizophrenia. In line with this, we hope to stimulate the development of a, to say it briefly, “Computational Systems Neuropsychiatry”.