PREDICTION OF RESIDUE EXPOSURE AND CONTACT NUMBER FOR SIMPLIFIED HP LATTICE MODEL PROTEINS USING LEARNING CLASSIFIER SYSTEMS

MICHAEL STOUT, JAUME BACARDIT, JONATHAN D. HIRST, JACEK BLAZEWICZ, NATALIO KRASNOGOR
2006 Applied Artificial Intelligence  
The performance of a Learning Classifier System (LCS) applied to the classification of simplified hydrophobic/polar (HP) lattice model proteins was compared to other machine learning (ML) algorithms. The GAssist LCS classified functional HP model proteins on the 3D diamond lattice as folding or non-folding at 88.3% accuracy, outperforming significantly three out of the four other methods. GAssist correctly classified HP model protein instances on the basis of Contact Number (CN) and Residue
more » ... sure (RE) on both 2D square and 3D cubic lattices at a level of between 27.8% and 80.9%. Again, the LCS performed at a level comparable to the other ML technologies in this task outperforming significantly them in 24 out of 180 cases, and being outperformed just six times. The benefits of using LCS for this problem domain are discussed and examples of the LCS generated rules are described.
doi:10.1142/9789812774118_0085 fatcat:vhlm2n3w4rc6tp72g7me3femhy