For many brain diseases, particularly in psychiatry, we lack clinical tests for differential diagnosis and cannot predict optimal treatment for individual patients. This presentation outlines a translational neuromodeling framework for inferring subject-specific mechanisms of brain disease from non-invasive measures of behaviour and neuronal activity. Guided by clinical theories of maladaptive cognition and aberrant brain-body interactions, generative models can be developed that have potential as “computational assays”. Evaluating the clinical utility of these assays requires prospective patient studies that address concrete clinical problems, such as treatment response prediction. If successful, computational assays may help provide a formal basis for differential diagnosis and treatment predictions in individual patients and, ultimately, facilitate the construction of mechanistically interpretable disease classifications.