Periodic Reporting for period 1 - NMDAR-DCM (Profiling NMDA receptor in schizophrenia and predicting clinical trajectories from rsEEG using dynamic causal modelling)
Reporting period: 2022-09-01 to 2024-08-31
Neuroscience aims to understand how the brain's fixed structure supports flexible neural dynamics crucial for adaptive behavior, mainly through synaptic transmission. N-methyl-D-aspartate receptors (NMDA-Rs), key ionotropic glutamate receptors, play a vital role in synaptic plasticity and are linked to learning, memory, and certain neuropsychiatric disorders like schizophrenia.
Currently, non-invasive methods to manipulate synaptic transmission are lacking. However, advances in dynamic causal models (DCMs) for electroencephalography (EEG) offer a potential solution by simulating neural activity and predicting synaptic dynamics.
Research using DCMs has shown promise in estimating NMDA-R parameters from resting-state EEG data, but its robustness and generalizability still need to be determined. This project seeks to enhance NMDA-R DCMs to improve their reliability and applicability across various EEG datasets, aiming for deeper insights into NMDA-R function and better interpretability in clinical and research settings.
The combination of a unique dataset of a large cohort of schizophrenic patients and the same number of controls in a prospective study design offers a unique opportunity to profile NMDA-R dysfunctions related to schizophrenia and tentatively predict the individual clinical trajectories that might allow for personalized treatment of patients by guiding selection of antipsychotic drugs with differential affinity to NMDA-Rs.
The specific scientific goals of this project are two-fold: methodological—developing and validating novel and robust DCM variants that are able to profile NMDA-R dynamics—and clinical—evaluating and possibly predicting clinical trajectories of schizophrenic patients. The methodological aim can be summarized with two goals:
1. Develop novel formulations of NMDA-R DCM applicable to rsEEG data and validate their robustness on within-subject animal data with pharmacological intervention via ketamine.
2. Cross-validate novel NMDA-R DCM on human resting-state EEG data using a within-subject design with pharmacological intervention via ketamine.
The final goal of this project then represents the clinical aim:
3. Perform NMDA-R profiling employing a novel DCM in schizophrenic patients and study its predictive power using generative embedding techniques.
This project will also help establish computational assays as a standard procedure in clinical practice, thereby making a step towards personalized mental health care in general.
Currently, non-invasive methods to manipulate synaptic transmission are lacking. However, advances in dynamic causal models (DCMs) for electroencephalography (EEG) offer a potential solution by simulating neural activity and predicting synaptic dynamics.
Research using DCMs has shown promise in estimating NMDA-R parameters from resting-state EEG data, but its robustness and generalizability still need to be determined. This project seeks to enhance NMDA-R DCMs to improve their reliability and applicability across various EEG datasets, aiming for deeper insights into NMDA-R function and better interpretability in clinical and research settings.
The combination of a unique dataset of a large cohort of schizophrenic patients and the same number of controls in a prospective study design offers a unique opportunity to profile NMDA-R dysfunctions related to schizophrenia and tentatively predict the individual clinical trajectories that might allow for personalized treatment of patients by guiding selection of antipsychotic drugs with differential affinity to NMDA-Rs.
The specific scientific goals of this project are two-fold: methodological—developing and validating novel and robust DCM variants that are able to profile NMDA-R dynamics—and clinical—evaluating and possibly predicting clinical trajectories of schizophrenic patients. The methodological aim can be summarized with two goals:
1. Develop novel formulations of NMDA-R DCM applicable to rsEEG data and validate their robustness on within-subject animal data with pharmacological intervention via ketamine.
2. Cross-validate novel NMDA-R DCM on human resting-state EEG data using a within-subject design with pharmacological intervention via ketamine.
The final goal of this project then represents the clinical aim:
3. Perform NMDA-R profiling employing a novel DCM in schizophrenic patients and study its predictive power using generative embedding techniques.
This project will also help establish computational assays as a standard procedure in clinical practice, thereby making a step towards personalized mental health care in general.
The first step in this project was to gain deep insights into resting state EEG data and their differential statistical properties based on the condition and status of the subject (e.g. healthy volunteer vs. schizophrenic patient). Based on the previous endeavors, we dove deep into EEG microstates, which are brief, stable patterns of brain activity that are thought to capture the dynamic organization of neural networks. These microstates, lasting only milliseconds, represent fundamental units of brain function and are critical for understanding resting-state EEG. By analyzing microstates, we can detect specific neural signatures and explore how these differ between healthy individuals and patients with schizophrenia. Research suggests that schizophrenia is associated with specific changes to the microstate properties, reflecting underlying disruptions in brain network connectivity. To capture these subtle, disorder-related changes, we investigated microstate dynamics. This foundational analysis is crucial, enabling us to identify potential resting state EEG markers we employ for a data-driven Bayesian model inversion using the DCM framework. We published this groundwork in the renowned journal NeuroImage.
Model inference relies on the stability and reliability of models interpreting brain activity, which is essential for sensitivity analysis, parameter recovery, and robustness evaluation. Sensitivity analysis assesses how small parameter changes affect output, guiding model refinement. Parameter recovery tests the accuracy of estimated values from observed data, ensuring outputs reflect neurobiological mechanisms. Robustness testing evaluates a model's stability across different datasets and conditions. Together, these analyses ensure meaningful insights into neural processes and increase confidence in using computational models to inform brain function and dynamics theories. We thoroughly researched, implemented, tested, and performed in-depth sensitivity and parameter recovery analyses using different methods. Analyses were done in a loop where we adapted the mathematical model definition, performed the analyses again, and compared the results. The results from our analyses were presented at the world-leading conference Organization for Human Brain Mapping Annual Meeting in 2023 and 2024.
Model inference relies on the stability and reliability of models interpreting brain activity, which is essential for sensitivity analysis, parameter recovery, and robustness evaluation. Sensitivity analysis assesses how small parameter changes affect output, guiding model refinement. Parameter recovery tests the accuracy of estimated values from observed data, ensuring outputs reflect neurobiological mechanisms. Robustness testing evaluates a model's stability across different datasets and conditions. Together, these analyses ensure meaningful insights into neural processes and increase confidence in using computational models to inform brain function and dynamics theories. We thoroughly researched, implemented, tested, and performed in-depth sensitivity and parameter recovery analyses using different methods. Analyses were done in a loop where we adapted the mathematical model definition, performed the analyses again, and compared the results. The results from our analyses were presented at the world-leading conference Organization for Human Brain Mapping Annual Meeting in 2023 and 2024.
The first result relates to the in-depth investigation of EEG microstate properties and dynamics and the understanding of how the (dysfunctional) brain dynamics shape the acquired EEG data. Our study was the first to systematically compare various algorithms for EEG data analysis, providing deep insights into the clustering and static and dynamic properties of such dimensionality reduction techniques. Our study also showed that many of these dynamical properties of EEG are determined purely by the linear structure of the EEG data itself, that is, its covariance and autocorrelation structure. To the best of our knowledge, our study is the most thorough analysis of the EEG dimensionality reduction technique that ties the EEG data properties, that can be used as data features in the model inversion, with the brain dynamics to this date.
The second novel result is more in the light of a technical note and recommendation for the practical use of Dynamical Causal Models, mainly related to their inference capabilities. It has been shown before that DCMs for M/EEG data might suffer from complexity, and their generalizability and robustness still need to be fully characterized. During the project, we systematically researched, implemented, tested, and utilized various mathematical and statistical techniques to inform the modeler of such model properties. To the best of our knowledge, this is separate from the typical repertoire for practical usage of DCMs in clinical settings. In return, the questionable robustness and generalizability hinder the potential clinical benefits of computational assays. Although we have not published our results as a journal article, we presented these findings and recommendations at two neuroscientific conferences, and our work received considerable interest from theoretical and clinical researchers. Once we publish our findings in a peer-reviewed scientific journal, our work will significantly help establish sensitivity and robustness analyses as a much-needed part of the clinical use of models stemming from computational neuroscience.
The second novel result is more in the light of a technical note and recommendation for the practical use of Dynamical Causal Models, mainly related to their inference capabilities. It has been shown before that DCMs for M/EEG data might suffer from complexity, and their generalizability and robustness still need to be fully characterized. During the project, we systematically researched, implemented, tested, and utilized various mathematical and statistical techniques to inform the modeler of such model properties. To the best of our knowledge, this is separate from the typical repertoire for practical usage of DCMs in clinical settings. In return, the questionable robustness and generalizability hinder the potential clinical benefits of computational assays. Although we have not published our results as a journal article, we presented these findings and recommendations at two neuroscientific conferences, and our work received considerable interest from theoretical and clinical researchers. Once we publish our findings in a peer-reviewed scientific journal, our work will significantly help establish sensitivity and robustness analyses as a much-needed part of the clinical use of models stemming from computational neuroscience.