Missense variants are the most abundant type of protein-altering variants in coding regions in the genome. Recent exome and genome sequencing studies have shown that missense variants overall contribute to half of attributable risk of autism among coding de novo variants. However, the genetic effect of individual missense variants is largely uncertain. At present, large-scale assessment of such variants relies mostly on computational algorithms based on conservation. Although these in silico prediction tools can serve as helpful guides for future laboratory experiments, they are not reliable for clinical diagnosis. Most missense variants found in genomic screens are currently designated as variants of uncertain significance. The uncertainty hinders the pace of risk gene discovery and the yield in genetic testing of autism and associated conditions.
There are opportunities now to use cutting-edge functional genomic and machine learning methods to provide better and clinically relevant interpretations for missense variants. New deep mutational scanning (DMS) paradigms are able to generate functional data on hundreds to thousands of protein variants in a single experiment1–2, providing novel insights into biologic mechanism and data for deeper modeling of protein function/pathogenicity3. Finally, a revolution is currently underway in structural biology that leverages deep-learning approaches to protein structure prediction with physics-based simulation algorithms to provide atomic level understanding of variant effects on a protein’s function4–5.
To fill this major gap and gain fundamental insights into the contribution of missense variants in autism, this study will develop: (1) new empirical DMS data of missense variants in transcription factor genes associated with autism risk, (2) advanced machine-learning models to predict functional impact and mode of action of missense variants and (3) biophysical models to interpret the functional impact of missense variants through the perturbation of protein energy landscapes (i.e., folding).
By triangulating these diverse data rich approaches, this study is expected to substantially improve the interpretation of missense variants in autism risk genes. Such information will be particularly valuable for increasing the diagnostic power of genetic testing for autism and related conditions. It will also help advance our understanding of the mode of actions of variants in these risk genes and will help spur the design of rationale treatment approaches.