DDGemb

DDGemb predicts changes in protein stability (ΔΔG, change in Gibbs free energy) resulting from single- and multi-point amino acid substitutions using protein language model embeddings and transformer architectures.

Key Features:

• Protein language model embeddings: Uses embeddings from protein language models to capture sequence–structure relationships relevant to stability.
• Transformer architectures: Employs transformer architectures to process embeddings and learn mapping from sequence variation to ΔΔG.
• Curated training dataset: Trained on a high-quality, literature-derived dataset curated for stability variation examples.
• Benchmark performance: Demonstrates state-of-the-art performance on benchmark datasets for both single- and multi-point variations.

Scientific Applications:

• Functional protein design: Inform design choices by predicting stability effects of candidate sequence variants.
• Protein engineering: Support engineering of proteins with desired stability properties for industrial or therapeutic applications.
• Disease mechanism elucidation: Provide hypotheses on how specific mutations alter protein stability and contribute to disease.

Methodology:

Trains on a curated, literature-derived dataset and applies protein language model embeddings with transformer-based machine learning to predict ΔΔG values for sequence variants.