Identifying transdiagnostic neurobiomarkers for precision major psychiatric disorders <i>via</i> machine learning: Causes, findings, and future directions

Abstract The diagnostic criteria of psychiatric disorders rest solely on symptoms rather than etiologies; this led to the lack of objective biomarkers for diagnosis, differential diagnosis and treatment of major psychiatric disorders (MPDs) including schizophrenia, bipolar disorder, and major depressive disorder. Facing this bottleneck problem, we firstly started our review from the evolution of psychiatric nosology for the diagnosis of MPDs which refined from categorical to dimensional classification. Specifically, the neuroimaging was sought as an intermediate phenotype to solve the dilemma introduced from a symptom-based diagnostic classification framework. Secondly, we reviewed findings applying traditional mass-univariate methods as well as machine learning methods to identify the diagnostic and treatment neurobiomarkers for MPDs from group level to individual level, and suggested the frontal-posterior functional imbalanced pattern as a potential transdiagnostic neurobiomarker for MPDs. We further initialized the necessity of using unsupervised learning approach for MPDs subtyping prior to performing supervised learning for classification, as seeking one single neurobiological feature should be incapable of coping with the strong heterogeneity in MPDs. We concluded that the ultimate goal of MPDs subtyping is to identify an objective neurobiomarker to guide clinical practice; based on frontal-posterior functional imbalance, we developed a subtyping and precise neuromodulation strategy to achieve the establishment of a precision medicine framework. Future studies should perform more etiological investigations based on this neurobiomarker, and conduct randomized controlled trials in larger MPDs populations for further verification.