An Enhanced Deep Learning Model for Efficient and Accurate Anti-Cancer Peptide Identification Surpassing Traditional Machine Learning Methods

Anti-cancer peptide (ACP) classification is crucial for developing new cancer therapies. However, traditional machine learning methods often struggle with both accuracy and efficiency in the task of ACP classification. To address these challenges, we utilized DistilBERT, which is a variant of BERT designed to reduce size and computational requirements while maintaining performance, to capture complex relationships and patterns in ACP sequences, by leveraging its ability to learn deep language features from large amounts of unlabeled data. DistilBERT’s knowledge distillation approach compresses the large BERT model into a smaller one, thereby reducing storage needs and accelerating inference speed. Our results demonstrate that DistilBERT significantly outperforms traditional machine learning models in ACP classification tasks, achieving an accuracy of 0.8957, precision of 0.9068, recall of 0.9012, and $F 1$-scores of 0.9045. This improved performance, combined with DistilBERT’s efficiency, makes it a promising tool for ACP classification, potentially accelerating the discovery and development of new cancer treatments, even in environments with limited computational resources.