AllesTM

AllesTM predicts multiple structural features of transmembrane proteins from atomic coordinate data to support structural and functional analysis.

Key Features:

• Integrated Machine Learning Framework: Employs Random Forests, Gradient Boosting Machines, convolutional neural networks (including dilated convolutions and residual connections), and long short-term memory (LSTM) architectures.
• Multi-Target Prediction: Simultaneously predicts residue depth in the membrane, flexibility, topology, relative solvent accessibility (bound and unbound states), torsion angles, and secondary structure.
• Automated Feature Engineering: Uses deep learning–facilitated automated feature engineering to support multi-target prediction.
• Performance: Demonstrates superior performance over existing methods for residue depth in the membrane, flexibility, topology, and bound-state relative solvent accessibility, and shows comparable performance to SPOT-1D for torsion angles, secondary structure, and monomer relative solvent accessibility.
• Algorithmic Effect Analysis: Provides analysis of the impact of different machine learning algorithms and parameter choices on prediction performance.

Scientific Applications:

• Protein function inference: Enables structural feature-based interpretation of transmembrane protein function.
• Membrane protein therapeutics design: Supports design and optimization of membrane protein–targeted therapeutics by providing structural feature predictions.
• Membrane protein interactions: Facilitates study of protein–protein interactions within cellular membranes via predicted topology, accessibility, and interface-related features.

Methodology:

Processes atomic coordinate data using Random Forests, Gradient Boosting Machines, convolutional neural networks (with dilated convolutions and residual connections), and LSTM architectures, with automated feature engineering enabled by deep learning.