Using the excitation/inhibition ratio to optimize the classification of autism and schizophrenia

Abstract The excitation/inhibition (E/I) ratio has been shown to be imbalanced in individuals diagnosed with autism (AT) or schizophrenia (SZ), relative to neurotypically developed controls (TD). However, the degree of E/I imbalance overlap between SZ and AT has not been extensively compared. In thi...

Full description

Saved in:
Bibliographic Details
Main Authors: Lavinia Carmen Uscătescu, Christopher J. Hyatt, Jack Dunn, Martin Kronbichler, Vince Calhoun, Silvia Corbera, Kevin Pelphrey, Brian Pittman, Godfrey Pearlson, Michal Assaf
Format: Article
Language:English
Published: Nature Publishing Group 2025-07-01
Series:Translational Psychiatry
Online Access:https://doi.org/10.1038/s41398-025-03455-8
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Abstract The excitation/inhibition (E/I) ratio has been shown to be imbalanced in individuals diagnosed with autism (AT) or schizophrenia (SZ), relative to neurotypically developed controls (TD). However, the degree of E/I imbalance overlap between SZ and AT has not been extensively compared. In this project, we used resting state fMRI data to estimate the E/I ratio via the Hurst (H) exponent. Our main objectives were (1) to quantify group differences in the E/I ratio between TD, AT, and SZ, (2) to assess the potential of the E/I ratio for differential diagnosis, and (3) to verify the replicability of our findings in an independently acquired dataset. For each participant, we computed the Hurst exponent (H), an indicator of the E/I ratio, from the time courses of 53 independent components. Using a random forest classifier (RF), we ran a classification analysis using the larger of the two datasets (exploratory dataset; 519 TD, 200 AT, 355 SZ) to determine which of the 53 H would yield the highest performance in classifying SZ and AT. Next, taking the ten most important H based on the exploratory dataset, in combination with phenotypic information collected in the replication dataset (55 TD, 30 AT, 39 SZ), we used RF to compare the classification performance using five feature sets: (a) H only; (b) Positive and Negative Syndrome Scale (PANSS) and the Autism Diagnostic Observation Schedule (ADOS) only; (c) PANSS, ADOS, Bermond–Vorst Alexithymia Questionnaire (BVAQ), Empathy Quotient (EQ), and IQ; (d) H, PANSS and ADOS; (e) H, PANSS, ADOS, BVAQ, EQ and IQ. Classification performance using H only was higher in the exploratory dataset (AUC = 84%) compared to the replication dataset (AUC = 72%). In the replication dataset, the highest classification performance was obtained when combining H with PANSS, ADOS, BVAQ, EQ and IQ (i.e., model e; AUC = 83%). These results illustrate the added value of E/I to typical phenotypic data in differentiating AT and SZ.
ISSN:2158-3188