Beyond Tripeptides Two-Step Active Machine Learning for Very Large Data sets.

Area of Science:

• Biomaterials Science
• Nanotechnology
• Computational Chemistry

Background:

• Self-assembling peptide nanostructures are crucial in nature and medicine.
• Previous studies screened di- and tripeptides, but higher-order combinations are computationally infeasible.
• Exhaustive simulation of complex peptide combinations presents significant challenges.

Purpose of the Study:

• To develop an efficient machine-learning method for discovering self-assembling peptide nanostructures.
• To overcome the computational limitations of screening large peptide sequence spaces.
• To enable the bottom-up manufacture of functional nanomaterials.

Main Methods:

• Developed an active machine-learning (iterative learning/evolutionary search) method.
• Utilized a low-resolution dataset for broad search space coverage.
• Employed a just-in-time high-resolution dataset for targeted analysis.
• Incorporated data curation (e.g., log P) to control candidate selection.

Main Results:

• The active machine-learning model efficiently searches vast peptide sequence spaces.
• The iterative approach improves both low- and high-resolution models with new data.
• The method allows for computationally efficient identification of ideal peptide candidates.
• Data curation strategies influence the selection of target peptides.

Conclusions:

• The developed active machine-learning method offers a computationally efficient approach to discovering self-assembling peptide nanostructures.
• This strategy overcomes the infeasibility of exhaustive screening for complex peptide combinations.
• The model is broadly applicable to other search spaces with minor algorithmic adjustments.