Designing Anticancer Peptides by Constructive Machine Learning

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Constructive machine learning (ML) allows automated generation of novel chemical structures.
• This approach bypasses the need for explicit molecular design rules.
• Peptides are promising therapeutic agents, but designing them de novo is challenging.

Purpose of the Study:

• To experimentally apply a deep ML model for de novo design of membranolytic anticancer peptides (ACPs).
• To evaluate the efficacy and selectivity of ML-generated ACP candidates.

Main Methods:

• A recurrent neural network with long short-term memory cells was trained on peptide sequences.
• Transfer learning was employed using 26 known ACPs to fine-tune the model.
• The optimized model generated novel amino acid sequences.
• Synthesized peptides were tested for activity against MCF7 cancer cells and selectivity against human erythrocytes.

Main Results:

• Ten of the twelve synthesized peptides exhibited activity against MCF7 cancer cells.
• Six of the active peptides demonstrated at least threefold selectivity, killing cancer cells while sparing erythrocytes.
• The generated peptides were novel and unique.

Conclusions:

• Constructive machine learning is effective for the automated design of peptides with specific biological activities.
• This AI-driven approach accelerates the discovery of potent and selective anticancer peptides.
• The study validates the use of deep learning for de novo peptide design in drug discovery.