Detection of HIV-1 progression phases from transcriptional profiles in ex vivo CD4+and CD8+T cells using meta-heuristic supported artificial neural network

Gene expression studies can explore various information about different diseases more specifically the working mechanism of diseases. Fortunately, HIV (Human Immunodeficiency Virus)-1 has a well-known pattern of progression. There are three stages through which HIV-1 progresses namely acute, chronic, and non-progressor. The detection of different stages is very important because late detection can lead to AIDS. Automated frameworks can be helpful in the precise detection of different progression stages. In this work, an automated framework has been proposed to detect several stages of HIV-1 progression. This work is based on the analysis of transcriptional profiles of CD4+ and CD8+ T cells. The microarray array data has been processed and reduced before classification. The detection process is based on the Artificial Neural Network which is trained with the help of meta-heuristic algorithms for better convergence. Three different metaheuristic algorithms namely GA, CPSO, and NSGA-II have been compared. The experimental results show that the artificial neural network trained with the Genetic Algorithm achieves 72.22% accuracy, 69.05% precision, 70.73% recall, and 69.88% F-Measure whereas the artificial neural network trained with the Constrained Particle Swarm Optimization achieves 86.67% accuracy, 78.79% precision, 83.87% recall, and 81.25% F-Measure. In contrast, the proposed approach i.e., the artificial neural network trained with the NSGA-II approach achieves 88.24% accuracy, 82.56% precision, 88.87% recall, and 85.6% F-Measure values and outperforms other approaches including decision tree, SVM, and KNN. The results have been verified using the cross-validation procedure that ensures and reflects the usefulness of the method.

Automated summary

This article proposes an automated framework based on transcriptional profile analysis to detect progression stages of HIV-1. By training an artificial neural network with metaheuristic algorithms, high accuracy and recall rates are achieved, outperforming other methods.