Imaging genetics and autism
© NeuroSpin/CEA.
This research involves using neuroimaging techniques genetics to study psychiatric and neurodevelopmental disorders, with a particular interest in autism spectrum disorder (ASD). ASD is a set of neurodevelopmental disorders defined by social difficulties, communicative deficits and restricted behaviours and interests, and typically affects 1-2% of the population. The heterogeneity of ASD makes it quite difficult to delineate its etiology and to reach a clear diagnosis, thus warranting the need for further characterization of the disorder.
With the use of machine learning methods, various methodological approaches will be applied to characterize ASD into specified subtypes based on three modalities: cognitive assessments, genetic SNPs and neuroimaging features. This will be done using the Healthy Brain Network (HBN) Cohort and will entail three major steps, the first two of which will be conducted on each modality. First, an unsupervised approach will be done using multiple correspondance analyses to uncover hidden patterns and trends. Next, based on what was ‘learned’ from the original data in the previous step, these hidden patterns will help us decide which features to include in the supervised learning model in order to develop an algorithm which will accurately predict ASD severity from new input data. Lastly, these methodologies will be implemented to robustly determine a stratification of ASD subtypes using a multi-block approach.
This research is important since it will contribute to the accurate characterization of ASD by distinguishing subtypes for more specialized, and eventually personalized, forms of treatment for individuals suffering from ASD. It will also lay the groundwork on expanding diagnosis criteria to include quantitative biological information and traits, thus differing from the current practice focusing primarily on qualitative questionnaire responses.