AllesTM: predicting multiple structural features of transmembrane proteins.

Area of Science:

• Structural bioinformatics
• Computational biology
• Protein structure prediction

Background:

• Transmembrane (TM) proteins possess unique physico-chemical properties distinct from globular proteins.
• Specialized prediction techniques are needed for TM proteins, but many structural features lack dedicated predictors.
• Deep learning facilitates automated feature engineering for multi-target prediction methods.

Purpose of the Study:

• To develop an integrated tool, AllesTM, for predicting diverse structural features of transmembrane proteins.
• To leverage multiple machine learning algorithms for enhanced prediction accuracy.
• To provide a user-friendly, comprehensive solution for TM protein structural bioinformatics.

Main Methods:

• Integration of various machine learning algorithms including random forests, gradient boosting machines, convolutional neural networks (CNNs), dilated CNNs, residual connections, and long short-term memory (LSTM) architectures.
• Prediction of multiple structural features from atomic coordinate data.
• Comparative analysis against existing methods like SPOT-1D.

Main Results:

• AllesTM demonstrates superior performance in predicting residue depth, flexibility, topology, and bound-state relative solvent accessibility.
• Prediction accuracy for torsion angles, secondary structure, and monomer relative solvent accessibility is comparable to leading methods.
• High accuracy across numerous prediction targets and ease of installation.

Conclusions:

• AllesTM offers a highly accurate and integrated prediction method for transmembrane protein structural features.
• The tool simplifies the process by reducing the need for multiple software installations.
• Provides insights into the performance of different machine learning algorithms for TM protein prediction.