Identifying and validating subtypes within major psychiatric disorders based on frontal–posterior functional imbalance via deep learning

Abstract

Converging evidence increasingly implicates shared etiologic and pathophysiological characteristics among major psychiatric disorders (MPDs), such as schizophrenia (SZ), bipolar disorder (BD), and major depressive disorder (MDD). Examining the neurobiology of the psychotic-affective spectrum may greatly advance biological determination of psychiatric diagnosis, which is critical for the development of more effective treatments. In this study, ensemble clustering was developed to identify subtypes within a trans-diagnostic sample of MPDs. Whole brain amplitude of low-frequency fluctuations (ALFF) was used to extract the low-dimensional features for clustering in a total of 944 participants: 581 psychiatric patients (193 with SZ, 171 with BD, and 217 with MDD) and 363 healthy controls (HC). We identified two subtypes with differentiating patterns of functional imbalance between frontal and posterior brain regions, as compared to HC: (1) Archetypal MPDs (60% of MPDs) had increased frontal and decreased posterior ALFF, and decreased cortical thickness and white matter integrity in multiple brain regions that were associated with increased polygenic risk scores and enriched risk gene expression in brain tissues; (2) Atypical MPDs (40% of MPDs) had decreased frontal and increased posterior ALFF with no associated alterations in validity measures. Medicated Archetypal MPDs had lower symptom severity than their unmedicated counterparts; whereas medicated and unmedicated Atypical MPDs had no differences in symptom scores. Our findings suggest that frontal versus posterior functional imbalance as measured by ALFF is a novel putative trans-diagnostic biomarker differentiating subtypes of MPDs that could have implications for precision medicine.

Response to: Significance and stability of deep learning-based identification of subtypes within major psychiatric disorders. Molecular Psychiatry (2022)

“…We also examined the effects of medication on symptom severity (from clinical data) to elucidate pharmacologic effects within each subtype. These validation analyses showed significant alignment between the subtyping results and these other biological features [2]. Much effort has been devoted to developing null models for clustering, as discussed in the whole Chapter 4 of the book by Bezdek [7].…”

“…

TO THE EDITOR:Recently, Winter and Hahn [1] commented on our work on identifying subtypes of major psychiatry disorders (MPDs) based on neurobiological features using machine learning [2]. They questioned the generalizability of our methods and the statistical significance, stability, and overfitting of the results, and proposed a pipeline for disease subtyping.

…”

The colors of our brain: an integrated approach for dimensionality reduction and explainability in fMRI through color coding (i-ECO)

“…While the current study supports the usage of i-ECO to classify fMRI participants according to diagnostic groups, considering previous discussed views offered by the clinical setting, the potential of a dimensional approach seems warranted. According to previous research in fact, neuroimaging biomarkers may have the potential to find different correspondences of psychopathology (Kebets et al, 2019;McTeague et al, 2017), in order to arrive to a more specific definition of cornerstone symptoms, their biological correlates and overall classifications supported by experimental results (Chang et al, 2020;Iravani et al, 2021;Schilbach et al, 2015;Tokuda et al, 2018). An integrated approach to neuroimaging has the potential for direct implications in the treatment of mental suffering and psychiatric practice (Iravani et al, 2021;Price et al, 2018), through a coordination of theoretical models for general psychiatry, psychotherapy, and neuroimaginge.g.…”

Modern Views of Machine Learning for Precision Psychiatry

“…By stacking multiple layers of autoencoders, a deep autoencoder model is formed to discover more complicated and potentially nonlinear feature patterns. The deep autoencoder model has been applied to extract lowdimensional features from the amplitude of low-frequency fluctuations in fMRI [139]. Clustering analysis with the latent features uncovered by the deep autoencoder successfully identified two subtypes within major psychiatric disorders including schizophrenia, bipolar disorder, and major depressive disorder.…”