Structure optimization of an artificial neural filter detecting membrane-spanning amino acid sequences

Area of Science:

• Bioinformatics
• Computational Biology
• Protein Science

Background:

• Integral membrane proteins play crucial roles in cellular functions.
• Predicting transmembrane regions is vital for understanding protein structure and function.
• Existing methods like hydrophobicity analysis have limitations.

Purpose of the Study:

• To develop an artificial neural network for accurate prediction of transmembrane regions in human integral membrane proteins.
• To offer an alternative and complementary method to hydrophobicity analysis.
• To investigate the role of amino acid side-chain properties in encoding membrane protein sequences.

Main Methods:

• Development of an artificial neural network using the structure evolution algorithm.
• Systematical optimization of network architecture and parameters.
• Training the network using incomplete induction based on input-output relations.
• Encoding amino acid sequences using seven physicochemical side-chain properties.

Main Results:

• The artificial neural network achieved 92% accuracy in predicting membrane/nonmembrane transition regions.
• Geometric properties of side chains were found to be of minor importance.
• Polarity, refractivity, and hydrophobicity were identified as key properties for feature extraction.

Conclusions:

• Membrane transition regions in proteins are encoded as characteristic features within amino acid sequences.
• The developed neural network provides a highly accurate method for transmembrane region prediction.
• The structure evolution algorithm offers a robust approach for developing sequence-based filters.