Deep capsule neural network for identifying anticancer peptides using sequence to image transformation-based local embedded features

Le résumé fourni par la source

BACKGROUND: Globally, cancer is a major health issue that poses a significant threat to human health. Traditional treatments and laboratory-based methods have been extensively employed to treat cancer-affected cells. However, their high processing costs and side effects still limit their efficacy. In the past decade, significant developments in the field of anticancer peptides (ACPs) have shown a promising alternative for developing reliable cancer drugs with low side effects. RESULTS: In this paper, we presented an effective model, pACP-CapsNet, to accurately identify ACPs. The input sequences are converted into structural and localized substitution-based images using SMR and RECM. Subsequently, HOG, DWT, and CLBP-based transformations are applied to the obtained two-dimensional images to produce novel feature spaces, including RECM_DCT, DWT_SMR, HOG_SMR, and RECM_CLBP. These extracted descriptors are then serially integrated to handle the drawbacks of individual descriptors. Additionally, the shuffled frog leaping algorithm is utilized for selecting the high-ranked features from the integrated hybrid vector. Several deep learning models are trained using SFLA features, among which the Capsule Neural Network (CapsNet) achieved higher prediction rates. The proposed pACP-CapsNet obtained an accuracy of 97.0% and an AUC of 0.98 using training samples. Further validation reveals that pACP-CapsNet outperformed available models, demonstrating improvements of approximately 3% and 4% using the ACP240 and ACP740 test sets, respectively. CONCLUSIONS: The confirmed efficiency and stability of the pACP-CapsNet model underscore its potential as a valuable tool in academic research, drug diagnosis, and drug design.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.