PRECOGx

PRECOGx predicts GPCR signaling mechanisms and transducer couplings from sequence by encoding GPCR sequences with deep-learning protein language model embeddings (ESM1b) to infer functional and structural determinants.

Key Features:

• Predictive Modeling: Uses machine learning to predict interactions between GPCRs and transducers including G proteins and β-arrestins.
• ESM1b Embeddings: Encodes GPCR sequences using ESM1b protein embeddings as input features.
• Binding Data Integration: Integrates binding data from publicly available studies to enhance predictive accuracy.
• Comprehensive Coverage: Covers all classes of GPCRs for broad applicability across the human GPCRome.
• Dimensionality Reduction: Projects input sequences onto a low-dimensional space that captures essential features of the human GPCRome for reference and variant tracking.
• Interpretability: Provides attention maps and predicted intramolecular contacts to inspect sequence and structural determinants of coupling.
• Variant Impact Prediction: Predicts the impact of disease-associated variants (ClinVar) and alternative splice forms from healthy tissues (GTEX).
• Model Comparison: Reports improved predictive performance relative to the prior model PRECOG.
• Associated Publication: PMID: 35639758.

Scientific Applications:

• GPCR-transducer coupling prediction: Predicts G protein and β-arrestin coupling profiles from GPCR sequence data.
• Functional repertoire mapping: Maps functional properties across the human GPCRome and provides a reference framework for variant analysis.
• Variant interpretation: Assesses functional impact of ClinVar-annotated disease variants and GTEX-derived alternative splice forms.
• Structural determinant analysis: Identifies sequence and intramolecular contact features associated with coupling using attention-derived signals.

Methodology:

Encodes GPCR sequences with ESM1b protein embeddings, integrates public binding data, applies machine learning/deep-learning predictive models, projects embeddings into a low-dimensional GPCRome space, and uses attention maps and predicted intramolecular contacts for interpretability.