High accuracy prediction of human papillomavirus types by statistical chaos representation and reduced dimensional quantization

Abstract

HPV genotyping is a significant approach to provide better diagnosis, medical treatment, and prevention strategies for fighting with cervical cancers. Firstly, ChaosCentroid and ChaosFrequency feature extraction techniques were proposed for HPV genotype prediction from whole genomes. ChaosCentroid captures the structure of nucleotide subsequences in terms of centroid, while ChaosFrequency extracts the statistical distribution of the subsequences along genomes. For predicting systems, multi-layer perceptron, radial basis function, k-nearest neighbor, and fuzzy k-nearest neighbor techniques were deployed. The experimental results showed that all methods yielded the highest prediction performance among the results obtained from several compared methods. But time complexity of the proposed techniques was considerably lower than the comparative alignment method. Secondly, ChaosPoly feature extraction technique was subsequently proposed for HPV genotype prediction from partial coding sequences. For each sub-region, ChaosPoly gives the precedence to the distribution of dot patterns in the chaos game representation in a form of polynomial. The fuzzy k nearest neighbor technique was deployed for identifying the corresponding HPV genotypes. The results showed that ChaosPoly outperforms ChaosCentroid and ChaosFrequency.