The nosology of mental illness is currently under intense discussion (Hyman 2010). Many in the field believe that the notion of discrete, categorical mental disorders, as currently instantiated in the ICD-10 and DSM-IV, requires rethinking. These psychiatric diagnostic systems employ criteria that are derived from clinicians' observations, patient self-report, and course. Therefore, almost by definition, these diagnostic categories must be biologically heterogeneous and are unlikely to have unique sets of causal factors and neurobiological underpinnings. Even at the level of clinical symptoms and signs, dimensionality and comorbidity are pervasive (Krueger and Markon 2011). This problem is reflected in the heterogeneity of DSM categories. For example, the schizophrenia concept can be generated out of 23 different combinations of symptoms and phenomena (Schwarz, VanBeveren, et al. 2011). As a consequence, misdiagnosis is a common occurrence in the psychiatric field.

At the level of etiology, genetic complexity (i.e. pleiotropy and polygenicity) appears to be the rule, with large numbers of interacting variants combining to confer shared genetic liability to broad domains of symptomatically related disorders, such as schizophrenia and bipolar disorder (Purcell et al. 2009). Extant data converge to support a model wherein multiple sets of genetic variants predispose the development of clustered symptom domains that are common to multiple disorders, rather than to specific categorical disorders as defined by DSM-IV diagnostic criteria. While neuroimaging has undoubtedly been useful in showing functional and structural abnormalities tied to diagnoses, as well as to genetic and environmental risk factors, much of that work has been pursued within one classical diagnostic category only, in relatively small datasets, and lacks replication.