AugACP Framework Achieves 0.9272 F1-Score, Outperforming Baselines in Novel Anticancer Peptide Identification

The discovery of novel Anticancer Peptides (ACPs) is crucial for developing new cancer therapies, given their low toxicity and high specificity. However, traditional experimental identification of ACPs is labor-intensive and low-throughput, creating a significant bottleneck in drug discovery. While intelligent computational models offer a promising alternative for virtual screening, existing databases suffer from limited ACP samples. This scarcity often leads to predictive models overfitting the available labeled data, resulting in suboptimal generalization to unseen peptide sequences and hindering the efficient identification of new therapeutic candidates.

Researchers developed AugACP, a novel semi-supervised prediction framework to enhance Anticancer Peptide (ACP) identification. It integrates semantic knowledge into masked Transformer pretraining for bio-inspired ACP sequence augmentation, generating reliable supervised signals. AugACP then uses a semi-supervised learning strategy, selecting high-confidence pseudo-labels from augmented peptides to improve prediction performance and generalization. The framework was evaluated against state-of-the-art baseline methods to identify novel oligopeptides.