TMP-SSurface2

TMP-SSurface2 predicts the surface accessibility of transmembrane protein residues using deep learning based solely on primary protein sequences.

Key Features:

• Attention-Enhanced BiLSTM Architecture: The method employs an attention-enhanced Bidirectional Long Short Term Memory (BiLSTM) network to abstract latent information from protein sequences without feature engineering.
• Improved Prediction Accuracy: The predictor achieves a Pearson correlation coefficient (CC) of 0.66 on an independent test dataset, improving from 0.58 in the previous version.
• Sequence-Only Input: The method predicts transmembrane protein surface accessibility using only primary sequences without requiring structural information.

Scientific Applications:

• Transmembrane Protein Structure Modeling: The tool supports structure modeling of transmembrane proteins based on primary sequences.
• Drug Target Analysis: The predictor enables identification of transmembrane protein surfaces to understand structural biases for material-binding of drugs or other biological molecules.

Methodology:

The method uses an attention-enhanced Bidirectional Long Short Term Memory (BiLSTM) network to learn from protein sequences and predict surface accessibility of transmembrane protein residues.