Fig. 2: Predictive decomposition of autism symptoms.

A parsimony-inducing pattern-analysis algorithm was used to search through the array of questionnaire domains and extract the most informative subsets of domains for predicting symptom severity in individuals with autism. a Domain groups: Trajectories of the classifier weights of the Autism Diagnostic Interview-Revised (ADI-R) and Autism Diagnostic Observation Schedule (ADOS) domains are plotted on the y axis, while the parsimony constraint of the statistical models decreasing from left to right (here represented as the increasing number of domains automatically selected) is plotted on the x axis. The curves indicate changes in the subset of selected domains (i.e., the weight set not equal to zero), typically an inclusion. The color of each line shows in which model solution a specific questionnaire domain is included as relevant. For example, ADI-R social and communication domains were found as part of the first (most parsimonious) predictive model and are plotted in red. b Prediction accuracy: The middle panel retraces how prediction performance increases step by step as the identified domain subsets are added to the model. Each colored point represents a predictive model, including a specific number of selected domains. Two domains were sufficient for decent prediction performance at the single-subject level. These two domains predicted autism severity with 88.81% accuracy, while the model including every domain of the ADI-R and ADOS predicted autism severity with 96.81% accuracy. c Relative domain importance: Domain importance in the active weights is indicated as the parsimony constraint that becomes more lenient (left to right). This panel thus represents the relative importance of each domain (y axis) as more variables are included in the model (x axis, from left to right). In sum, the results emphasize that using every domain of the ADI-R and ADOS, autism severity was predicted with 97% accuracy; while using only the social and communication domains of the ADI-R, autism severity was predicted with an accuracy as high as 89%, indicating a very high predictive power for these two elements.