Deep Learning for Prediction and Optimization of Fast-Flow Peptide Synthesis

Area of Science:

• Chemical Synthesis
• Biotechnology
• Computational Chemistry

Background:

• Peptide synthesis relies on efficient amide bond formation.
• Sequence-dependent aggregation can reduce yields in solid-phase synthesis.
• Automated synthesizers require real-time monitoring for optimization.

Purpose of the Study:

• To apply deep learning to UV-vis data for analyzing peptide synthesis.
• To predict fluorenylmethyloxycarbonyl (Fmoc) deprotection efficiency.
• To minimize aggregation events during automated peptide synthesis.

Main Methods:

• Collected UV-vis analytical data from 35,427 Fmoc deprotection reactions.
• Utilized an automated fast-flow peptide synthesizer.
• Developed a deep learning model to map sequence data to synthesis parameters.

Main Results:

• The computational model predicted deprotection outcomes with less than 6% error.
• Analysis of UV-vis trace features (integral, height, width) correlated with coupling cycles.
• Demonstrated prediction of Fmoc deprotection efficiency and aggregation minimization.

Conclusions:

• Deep learning provides an effective method for analyzing real-time peptide synthesis data.
• This approach enables computationally designed optimization of peptide synthesis.
• Establishes a foundation for real-time optimization of peptide synthesis in flow.