Characterizing autism phenotypes by multimedia signal and natural language processing
Noémie Elhadad, Ph.D.
Partha Mitra, Ph.D.
Cold Spring Harbor Laboratory
A major challenge facing autism research is the lack of objective measures to characterize the disorder’s various phenotypes. More quantitative phenotypic measures, derived objectively from biological measurements, would facilitate genetic association studies by allowing biologically heterogeneous populations to be categorized into subgroups. They could also help refine diagnostic procedures.
Noémie Elhadad from Columbia University and Partha Mitra from Cold Spring Harbor Laboratory plan to address this issue using data from the Simons Simplex Collection (SSC), a cohort of families that have one child with autism and unaffected siblings and parents. The SSC includes video recordings of patients undergoing the Autism Diagnosis Observation Schedule (ADOS) protocol, a standardized interview for assessing and diagnosing autism.
The researchers plan to analyze these recordings using two techniques: natural language processing, which evaluates linguistic features of the subjects’ speech, such as intonation and use of unusual word combinations, and multimedia signal processing, which assesses audiovisual elements of the recordings such as the speakers’ facial emotions and where they are looking. They will then combine the findings with the subjects’ ADOS scores and other demographic and clinical features to develop a set of quantitative phenotypic measures for autism. The processed data will be added to the SSC knowledge base for use by other researchers. Elhadad and Mitra also plan to make their linguistic and audiovisual analysis platform publicly available.