Figure 1

Workflow overview and language sample instruction. (A) The workflow for this project starts with data collection via phone or in-person interviews. Five-minute language samples were generated in response to interview instruction shown in (B) as previously described¹¹. The audio recordings were transcribed and a text-processing module was used to extract linguistic features from the resulting transcripts. In parallel, cognitive profiles were collected using a cognitive self-assessment based on BRIEF-A standard^{32, 33}. Then, a comprehensive profile was generated using the combination of linguistic and cognitive features. For the machine learning portion of the study, a feature selection module was implemented to choose the most informative features based on the information gain score. A classifier was developed to determine the FX premutation status. Finally, the results of evaluation were used to determine the feasibility of improving data collection process. A mobile app based on these results was developed to expand the study.