The failure of proteins to fold correctly result in amyloidosis. Therefore, amyloid plaque prediction has become
significant to narrow down the exploration of anti- amyloidosis and related drugs. In this research article, we propose a
unique hybrid approach to computationally predict the formation of amyloid plaques by exploiting diversity in the feature
vector extracted from protein sequences and structures. The diversity in the sequence of feature space is exploited using
structure dependent features besides the physico-chemical information from amino acid chemistry and frequency spectrum
based parameters. We explored the prediction capability with independent and integrated feature vectors by an ensemble
machine learning classifier, Random Forests. Computational analysis evidence that the assimilation of diverse feature
set outperform individual feature array with a balanced prediction accuracy of 0.830 and Receiver Characteristic
Curve area of 0.918 on stratified10-fold cross-validation test.
Keywords: Amyloid plaque, fibrillogenesis, frequency spectrum parameters, physico-chemical properties, random forests,
structure dependent features.
Rights & PermissionsPrintExport