Reducing overfitting in predicting intrinsically unstructured proteins

Authors:
Pengfei Han;Xiuzhen Zhang;Raymond S. Norton;Zhiping Feng
Affiliations:
School of Computer Science and IT, RMIT University, Melbourne, VIC, Australia;School of Computer Science and IT, RMIT University, Melbourne, VIC, Australia;The Walter and Eliza Hall Institute of Medical Research, Parkville, VIC, Australia;The Walter and Eliza Hall Institute of Medical Research, Parkville, VIC, Australia
Venue:
PAKDD'07 Proceedings of the 11th Pacific-Asia conference on Advances in knowledge discovery and data mining
Year:
2007

Citing 5
Cited 0

C4.5: programs for machine learning

C4.5: programs for machine learning
Machine Learning

Machine Learning
Random Forests

Machine Learning
Prediction of unfolded segments in a protein sequence based on amino acid composition

Bioinformatics
Reducing overfitting in process model induction

ICML '05 Proceedings of the 22nd international conference on Machine learning

Quantified Score

Hi-index	0.00

Visualization

Abstract

Intrinsically unstructured or disordered proteins are proteins that lack fixed 3-D structure globally or contain long disordered regions. Predicting disordered regions has attracted significant research recently. In developing a decision tree based disordered region predictor, we note that many previous predictors applying 20 amino acid compositions as training parameter tend to overfit the data. In this paper we propose to alleviate overfitting in prediction of intrinsically unstructured proteins by reducing input parameters. We also compare this approach with the random forest model, which is inherently tolerant to overfitting. Our experiments suggest that reducing 20 amino acid compositions into 4 groups according to amino acid property can reduce the overfitting in decision tree model. Alternatively, ensemble-learning techniques like random forest is inherently more tolerant to this kind of overfitting and can be a promising candidate in disordered region prediction.