Clustering proteins into families using artificial neural networks

Area of Science:

• Bioinformatics
• Computational Biology
• Artificial Intelligence

Background:

• Protein family classification is crucial for understanding biological function and evolution.
• Existing methods for protein clustering can be limited by predefined parameters or computational intensity.

Purpose of the Study:

• To develop and evaluate an unsupervised artificial neural network for protein family clustering.
• To assess the network's ability to self-organize protein data and adapt clustering resolution.

Main Methods:

• Utilized a 7x7 artificial neural network trained with the Kohonen unsupervised learning algorithm.
• Input data consisted of matrix patterns derived from the bipeptide composition of 447 proteins across 13 families.
• A second experiment trained the network on a specific protein family (cytochrome c) to test adaptive resolution.

Main Results:

• The network successfully clustered 96.7% of proteins into their correct families without prior knowledge of family structure.
• Demonstrated self-organization into topologically ordered maps, reflecting protein relationships.
• The network adapted clustering resolution, grouping 76 cytochrome c sequences into 25 distinct neurons, with related sequences positioned adjacently.

Conclusions:

• The artificial neural network provides an effective and adaptable tool for unsupervised protein clustering.
• This approach facilitates rapid classification of new proteins and aids in homology searches in large macromolecular databases.
• The method shows promise for organizing and analyzing complex biological sequence data.