What is E/I imbalance? Clarifying the fundamental circuit dysfunction in schizophrenia using biophysical modelling of multiple imaging paradigms

Thursday, January 7, 2021 – 4:00 – 5:00 pm

Dr. Rick A Adams, MRC Fellow, Department of Computer Science & Max Planck UCL Centre for Computational Psychiatry

What?

Can biophysical circuit models inform on the fundamental circuit pathology in schizophrenia/ psychosis? If so, might these models be able to give us 'computational biomarkers' for pathology in individual patients? This presentation describes how dynamic causal modelling (DCM) can be employed to answer these questions by deriving subject-specific parameters from evoked responses, time-frequency spectra, resting fMRI, and other types of data. DCM could also be a very powerful tool for the purpose of characterizing circuit pathology in other disorders, e.g. autistic spectrum disorder.

Why?

Most computational models aim to simulate overall data features of interest but don't necessarily provide subject-specific parameters. In contrast, dynamic causal modelling can fit models to individual subjects' data, providing quantitative measures to each individual subject. DCM also has other advantages, in that the same framework contains numerous different kinds of biophysical models that can model e.g. evoked responses, time-frequency spectra, resting fMRI, all within the same framework
It is important to keep several potential DCM drawbacks in mind. First, the EEG models have a lot of parameters and thus can overfit or have difficulty converging. Care needs to be taken to reduce these problems. For fMRI this is not a

big issue. Second, results of modeling based on small samples have to be treated with caution - they are probably overconfident. Lastly, causal inferences cannot always be drawn from observational data, thus for maximum reliability, DCM has to be taken in tandem with other experimental approaches; if they all agree, one can be more confident in the results. That said, DCM is a very popular approach, and editors at key neuroimaging journals are happy with it if it's done well.

How?

Prerequisites: include some familiarity with biophysical models (for M/EEG work) and some basic programming skills in Matlab (requires a license) and SPM (free).

Reference publications:

• Stephan et al., 2009. “Ten simple rules for dynamic causal modeling.” ;
• For EEG: Moran et al., 2013.“Neural masses and fields in dynamic causal modeling”
• For fMRI and Parametric Empirical Bayes: Zeidman et al., 2019ab. “A guide to group effective connectivity analysis, part 1: First level analysis with DCM for fMRI”; “A guide to group effective connectivity analysis, part 2: Second level analysis with PEB”

Datasets: any preprocessed fMRI/EEG/MEG data can be used in theory, although the better-designed the experiment, the better the modelling will work. An online SPM manual (https://www.fil.ion.ucl.ac.uk/spm/) contains step by step instructions on analyzing an open dataset.

SPM on DCM tutorials online:

• fMRI/PET/VBM: Dynamic Causal Modelling for fMRI; Demo: DCM for fMRI; DCM for fMRI, Advanced Topics
• MEG/EEG: The principles of DCM; DCM for evoked responses; DCM for steady state responses; DCM for time-frequency; Demo: DCM for MEG/EEG; Demo: DCM for ERP/ERF

Registration Information

You must subscribe to the Workshop series using your email to receive Zoom information for the virtual workshop: https://medicine.yale.edu/psychiatry/map/