WP6: Predicting the individual’s response to tDCS
The main objectives of WP6 are:
- To develop multivariate biomarkers (e.g. utilizing EEG and/or neuroimaging measures) that accurately predict response to tDCS for individual cases.
- To create strong synergistic effects across WPs by translating outcome measures into clinically relevant constructs that can be used to inform complex clinical decision making.
- To stratify individuals with neurodevelopmental conditions (i.e. ASD, ADHD) based on their response to tDCS by examining differences in the neurobiological underpinnings between responders and nonresponders
In work package 6, the analysis of statistical effects is shifted from the group level to the individual (i.e. case) level in order to make predictions about which individuals will benefit the most from tDCS. To do so, conventional analytical techniques for making predictions at the individual level are combined with state-of-the-art multivariate pattern classification (MVPC) approaches that are particularly well suited for the prediction of binary or continuous outcome measures based on a set of complex, multivariate biological data. MVPC classifiers are initially 'trained' on a well-characterized sample of individuals with known class labels (e.g. responders vs. non-responders) in order to identify a complex pattern of data that can then be applied to new individuals to make a prediction. The training data will be provided by work package 3, work package 4, and work package 5, and will consist of neuroimaging and electrophysiological measures that will (i) be examined separately, and (ii) in combination (e.g. using multi-modal data-fusion techniques), to establish a pattern of biological features that best predicts response to tDCS. The insights into the mechanisms of clinical efficacy of tDCS provided by work package 7 will be implemented into clinically relevant constructs. The integrated approach to data analysis, which constitutes an essential part of WP6, is expected to provide more sensitive and accurate biomarkers, and a better understanding of the underlying mechanisms predicting and mediating response to tDCS than either modality alone.
The overall work conducted as part of work package 6 is subdivided into the following tasks:
- Prediction of response to tDCS in neurotypical controls (Task 1)
A biologically-driven predictive model of the individual's response to tCDS in neurotypical controls across different stages of development will be established. Using machine-learning approaches (e.g. Support Vector Machine, Gaussian Process Classification), the biological data acquired in work package 3 (i.e. anatomical MRI, resting-state fMRI, Diffusion Tensor Imaging, EEG) will be utilized, in combination with the known clinical outcomes of the tDCS, in order to develop a 'normative' classifier that will allow to distinguish between responders and non-responders based on their neurobiological and genetic make-up (i.e. stratification of individuals based on their responsiveness to tDCS). This will enable us to establish complex neurobiological markers (e.g. particular characteristics in the individual's neuroanatomical and/or genetic make-up) that are facilitative for response to tDCS. Moreover, response to tDCS will be predicted in a quantitative fashion (i.e. prediction of continuous outcome measures) using Gaussian-Progress Regression, which will result in probabilistic estimates (i.e. level of confidence) for the prediction of response to tDCS for individual cases. The biological plausibility of the model will be assessed based on conventional between-group comparisons of the individual biological features that most accurately predict response to tDCS. This work will be conducted in collaboration with work package 3 and work package 7.
- Prediction of response to tDCS in ADHD (Task 2)
The normative model established in Task 1 will be applied to predict the response to tDCS in individuals with ADHD. This will allow (i) to validate our biologically-based classifier based on an independent sample of individuals, and (ii) to examine whether the biological features that predict response to tDCS in neurotypical controls can also be utilized for the prediction of response to tDCS in individuals with ADHD. If it is found that the normative model does not provide accurate predictions for tDCS responses in ADHD, a new classifier that is optimized for predicting response to tDCS in this group of patients will be developed. Here, model validation will be performed within the available sample (e.g. using leave-one-out cross-validation). The biological plausibility of the ADHD classifier will be assessed based on the insights into the neurobiological underpinnings of ADHD examined in WP7. This work will be conducted in collaboration with work package 4 and work package 7.
- Prediction of response to tDCS in individuals with ASD (Task 3)
Here, our biological modeling approach will be applied to the individuals with ASD that were examined as part of work package 5. ASD is a neurodevelopmental condition with a large clinical and phenotypic heterogeneity that most likely results from a complex, multifactorial etiology. Moreover, ASD continues to be diagnosed on the behavioral level alone (i.e. based on symptomatology). Hence, clinical trials typically include samples of biologically heterogeneous individuals that are unlikely to respond well to the same (i.e. 'one-size-fits-all') approach to treatment and/or intervention. This most likely also includes tDCS. To establish the success of tDCS in the ASD group, it is therefore crucial to examine inter-individual variability in response to treatment, and to separate responders from non-responders, as these may have different neurobiological underpinnings. The data provided by work package 5 will therefore offer the unique opportunity to stratify ASD individuals based on their response to tDCS, and hence to disentangle the large phenotypic heterogeneity among ASD individuals by characterizing biologically more homogeneous subgroups. Moreover, the neurobiological and genetic markers that are characteristic for responding/non-responding ASD individuals will be implemented into a predictive model that can subsequently be used to inform the development of treatment strategies that are individually tailored, and that maximize treatment success within and across ASD individuals. This work will be conducted in collaboration with work package 5 and work package 7.