Multimodal genotype-phenotype analysis in SMARCB1-associated developmental disorders

Variants in SMARCB1, encoding a core subunit of the BAF (ie, BRG/BRM-associated factor) chromatin remodeling complex, are associated with intellectual developmental disorders, particularly Coffin-Siris syndrome, although the genotype-phenotype spectrum remains incompletely defined. This study aims to assess the correlations between SMARCB1 variant location and phenotypic manifestations.

We analyzed 31 individuals with pathogenic or likely pathogenic SMARCB1 variants using multimodal approaches, integrating clinical, structural, and machine learning analyses. We predicted variant effects via 3-dimensional protein modeling, assessed facial similarity using GestaltMatcher, and conducted phenotype-driven genotype prediction using machine learning classifiers.

Variants clustered within the N-terminal (winged-helix/SNF5) and C-terminal (αC-helix) regions. C-terminal Coffin-Siris syndrome variants were associated with more severe speech delay, microcephaly, and cleft palate, exhibiting stronger facial gestalt similarity. XGBoost achieved 96.7% accuracy in classifying variant location from phenotype alone. Although gestalt is a key feature delineating variants at the αC helix, overall clinical features have greater predictive power for N-terminal variants.

Using detailed phenotyping and machine learning algorithms, we identified differences between individuals with N-terminus and C-terminus SMARCB1 variants. Our study underscored the importance of multimodal assessments for genotype-phenotype associations, suggesting that integrated modeling can provide insights into SMARCB1 variant effects and biological function, with the potential for improvement in diagnostic strategies.