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

Area of Science:

• Computational biology
• Bioinformatics
• Drug discovery

Background:

• Cancer remains a significant global health challenge.
• Traditional cancer treatments face limitations due to high costs and side effects.
• Anticancer peptides (ACPs) offer a promising alternative for cancer therapy.

Purpose of the Study:

• To develop an effective computational model for accurate identification of ACPs.
• To leverage deep learning for enhanced prediction of anticancer peptide sequences.
• To address limitations of existing methods in ACP identification.

Main Methods:

• Input sequences were transformed into images using SMR and RECM.
• Feature extraction involved HOG, DWT, and CLBP transformations, creating hybrid feature spaces.
• The shuffled frog leaping algorithm (SFLA) was used for feature selection.
• Capsule Neural Network (CapsNet) was employed for classification.

Main Results:

• The pACP-CapsNet model achieved 97.0% accuracy and 0.98 AUC on training data.
• The model demonstrated superior performance over existing methods on ACP240 and ACP740 test sets.
• Integrated features and SFLA-selected features improved prediction rates.

Conclusions:

• The pACP-CapsNet model shows high efficiency and stability for ACP identification.
• This tool has potential applications in academic research and cancer drug design.
• The model facilitates drug diagnosis and development of novel cancer therapeutics.