ACP-ESM: A novel framework for classification of anticancer peptides using protein-oriented transformer approach

Area of Science:

• Biochemistry
• Computational Biology
• Oncology

Background:

• Anticancer peptides (ACPs) are promising therapeutic agents due to their selective cancer cell targeting.
• Challenges remain in developing accurate prediction models for ACP identification.

Purpose of the Study:

• To develop an efficient transformer-based framework for accurate prediction of anticancer peptides.
• To evaluate the performance of different transformer models (ESM, ProtBERT, BioBERT, SciBERT) for ACP identification.

Main Methods:

• Utilized four transformer models: ESM, ProtBERT, BioBERT, and SciBERT.
• Trained and tested models on widely-used ACP datasets: AntiCp2, cACP-DeepGram, and ACP-740.
• Developed an efficient transformer-based framework for ACP prediction.

Main Results:

• The proposed ESM framework achieved state-of-the-art accuracy: 96.45% on AntiCp2, 97.66% on cACP-DeepGram, and 88.51% on ACP-740.
• The framework demonstrated enhanced classification accuracy compared to existing literature studies.
• The developed model significantly improves the reliability and precision of ACP prediction.

Conclusions:

• The transformer-based framework, particularly the ESM model, offers a highly accurate and reliable method for identifying anticancer peptides.
• This advancement holds significant potential for accelerating the development of novel cancer therapies.
• The publicly available code facilitates further research and application in cancer drug discovery.