Bioinformatics: the machine learning approach
Bioinformatics: the machine learning approach
Hi-index | 0.00 |
Prediction of protein secondary structures from amino acid sequences is a useful intermediate step for further elucidation of native, three-dimensional conformation of proteins. Currently, most predictors are based on machine learning approaches with a short fixed-size input window scanning over the amino acid sequence. The center of the window corresponds to the prediction site where the prediction is performed by utilizing the properties of neighboring amino acid residues. By nature, most machine learning approaches consider feature vectors as position-independent in terms of feature components. As such, for the secondary structure prediction problem, most existing approaches do not take into account the distance of amino acid residues from the center residue. We have studied on how the prediction performance can be affected by imposing different weights on the features according to the distance of residues from the center residue, and in this work, we propose an adaptive weighting scheme to improve prediction accuracy.