Towards generalizable predictions for G protein-coupled receptor variant expression.

Area of Science:

• Biochemistry
• Genetics
• Computational Biology

Background:

• Missense mutations affecting integral membrane protein plasma membrane expression (PME) cause genetic diseases.
• Distinguishing mutations impacting PME from those altering protein activity is crucial for targeted therapeutics.
• Predicting mutation effects on membrane protein stability and expression remains a significant challenge.

Purpose of the Study:

• To develop and validate artificial neural networks for predicting the PME of transmembrane domain variants.
• To leverage deep mutational scanning data with structural and evolutionary features for accurate PME prediction.
• To differentiate pathogenic mutations causing misfolding from those affecting signaling in G protein-coupled receptors.

Main Methods:

• Utilized deep mutational scanning data to train artificial neural networks (ANNs).
• Employed structural and evolutionary features to predict PME of transmembrane domain variants.
• Tested ANN performance on rhodopsin, β2 adrenergic receptor, and KCNQ1 channel variants.

Main Results:

• The best-performing ANN, the PME predictor, accurately recapitulated mutagenic trends in rhodopsin.
• The PME predictor successfully differentiated pathogenic variants causing misfolding versus signaling defects.
• The network showed predictive power for another G protein-coupled receptor but not a voltage-gated potassium channel, suggesting fold-specific generalizability.

Conclusions:

• Structural features are largely sufficient for predicting mutagenic trends in membrane protein PME.
• ANNs trained on deep mutational scanning data show promise for predicting PME and may be generalizable across proteins with similar folds.
• Findings have implications for designing mechanistically specific genetic predictors for membrane protein diseases.