ProtT5-MSCRNet: a multi-scale convolutional and channel-recalibrated deep learning framework for anticancer peptide...

Area of Science:

• Biotechnology
• Computational Biology
• Drug Discovery

Background:

• Cancer is a major global health concern, driving the search for effective treatments.
• Anticancer peptides (ACPs) show promise due to their low toxicity and targeted action.
• Experimental identification of ACPs is inefficient, costly, and time-consuming.

Purpose of the Study:

• To develop an advanced computational framework for accurate Anticancer Peptide (ACP) prediction.
• To overcome the limitations of traditional experimental methods in ACP discovery.
• To provide a robust and interpretable tool for accelerating anticancer peptide research.

Main Methods:

• Developed ProtT5-MSCRNet, an end-to-end deep learning framework.
• Integrated ProtT5 evolutionary representations and multi-scale convolutional neural networks.
• Employed channel-wise attention recalibration and advanced optimization techniques.

Main Results:

• ProtT5-MSCRNet significantly outperformed existing state-of-the-art ACP prediction methods on two independent datasets.
• Achieved high performance metrics, including ACC, SN, SP, and MCC values exceeding 0.95 on Test Set 2.
• Ablation studies and visualization confirmed the model's effectiveness and interpretability.

Conclusions:

• ProtT5-MSCRNet is a powerful computational tool for identifying potential anticancer peptides.
• The framework facilitates and accelerates the discovery of novel ACPs for cancer therapy.
• Offers a biologically meaningful and accurate approach to ACP prediction.