Machine Learning to Accelerate the Discovery of Therapeutic Peptides
- 1
- 2Leibniz Institute of Plant Biochemistry
Open Access
closed
The discovery and design of therapeutic peptides have been significantly accelerated by the integration of machine learning approaches, which enable the prediction, generation, and optimization of bioactive sequences. This chapter provides a comprehensive overview of how machine learning-based strategies are transforming peptide science. We first introduce the biological relevance, advantages, and limitations of therapeutic peptides compared to traditional small-molecule drugs. Subsequently, we explore key computational methodologies for peptide discovery, including supervised learning pipelines for bioactivity classification, toxicity risk assessment, and pharmacological property prediction. Emphasis is placed on the use of advanced sequence encoding strategies, protein language models, and deep learning architectures. Recent innovations in generative learning, including variational autoencoders, generative adversarial networks, transformer-based models, and diffusion frameworks, are discussed as pivotal tools for the de novo design of peptides with tailored properties. Furthermore, the chapter highlights the emerging role of autonomous multi-agent systems that integrate generative and predictive models with reinforcement learning, offering adaptive platforms capable of iteratively improving peptide candidates based on experimental feedback. By combining computational power with experimental validation, these strategies are reshaping the landscape of therapeutic peptide engineering, enabling the rapid discovery of novel, personalized therapeutics with enhanced efficacy and safety profiles. Finally, we discuss the current challenges and future directions in the development of intelligent, data-driven pipelines for the next generation of therapeutics peptides.