FoldTR

FoldTR performs protein fold recognition by integrating triplet networks and ensemble deep learning to directly optimize protein fold embeddings for structure and function prediction.

Key Features:

• Triplet Network Integration (FoldNet): A deep convolutional neural network named FoldNet is trained with triplet loss to directly optimize protein fold embeddings so that proteins with similar folds are closer in embedding space.
• Residue-Residue Contact Assistance: A residue-residue contact-assisted predictor leverages the optimized FoldNet embeddings to enhance protein fold recognition accuracy.
• Ensemble Deep Learning Methodology (FSD_XGBoost): FSD_XGBoost integrates the FoldNet embedding with two additional discriminative fold-specific features extracted by SSAfold and DeepFR to produce complementary feature sets for classification.
• Improved Sensitivity: The FSD_XGBoost ensemble achieves a Top 1 sensitivity of 74.8%, approximately 9% higher than prior state-of-the-art methods.

Scientific Applications:

• Protein structure prediction: Provides more accurate fold-level assignments to support protein structure prediction.
• Functional annotation: Improves inference of protein function through enhanced fold recognition.
• Structural biology and drug discovery: Supports target characterization in structural biology and aids drug discovery efforts by improving fold recognition.

Methodology:

FoldTR combines triplet network strategies with ensemble deep learning: FoldNet directly optimizes protein fold embeddings via triplet loss, a residue-residue contact-assisted predictor uses these embeddings, and FSD_XGBoost integrates the embedding with SSAfold and DeepFR–derived features for classification.