A scalable reinforcement learning approach for screening large peptide libraries for bioactive peptide discovery

Area of Science:

• Biochemistry
• Computational Biology
• Drug Discovery

Background:

• Bioactive peptides, including anticancer peptides (ACPs), present a promising therapeutic strategy against cancer due to their tumor selectivity and low toxicity.
• Discovering novel ACPs is challenging due to the vast chemical space and computationally intensive in silico screening methods.

Purpose of the Study:

• To develop a cost-efficient computational method for exploring large peptide libraries using reinforcement learning and posterior sampling.
• To identify novel membranolytic peptides with therapeutic potential for cancer treatment.

Main Methods:

• Utilized reinforcement learning and posterior sampling for efficient exploration of peptide libraries.
• Screened a focused library of 36 million structurally resolved helical peptides from the Protein Data Bank.
• Applied in vitro assays to evaluate the cytotoxic activity of selected peptide candidates against cancer cells.

Main Results:

• The computational method reduced the search space by over 90% compared to exhaustive screening.
• 15 out of the top 100 selected peptide candidates showed cytotoxic activity against breast cancer cells, including triple-negative breast cancer.
• Three lead compounds were identified as non-toxic to healthy human cells.

Conclusions:

• Deep reinforcement learning can significantly expedite the discovery of bioactive peptides.
• This computational approach offers a promising pathway for developing novel peptide-based cancer therapies.
• The identified peptides demonstrate potential for treating various forms of breast cancer.