Novel weighted amino acid composition for prediction of protein structural classes within the context of multi-sensor data fusion approach

Prediction of structural classes of proteins is one of the most important but challenging research problems in computational biology and mainly based on amino acid sequence of the proteins. However, most of the predictive features based on the sequences donpsilat consider natural amino acid scales, which have been shown to play an important role in characterising the proteins in other studies. Therefore, this paper aims to present development of a novel weighted amino acid composition features based on the amino acid scales and multi-sensor data fusion strategies for reliable and accurate prediction of the structural classes of the proteins. The approach is further developed applying principal component analysis in each weighted amino acid composition features, which then leades to a locally optimized multi-sensor data fusion model. This pilot study presents promising results that show that the methods improve predictive accuracy by 1 to 10% compared to previously studied methods for the same data set. The approach taken is also shown to be not only effective, but also, in particular, more informative as it fuses information obtained from natural amino acid index scales that help better understand nature of such proteins.