Solution-Phase DNA-Compatible Pictet-Spengler Reaction Aided by Machine Learning Building Block Filtering.

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Synthetic Chemistry

Background:

• DNA-encoded library (DEL) technology is a powerful platform for drug discovery.
• The integration of machine learning (ML) with DEL synthesis remains underexplored.
• Predicting reaction success for building blocks is crucial for efficient DEL library design.

Purpose of the Study:

• To develop and validate a machine learning algorithm for predicting DNA-compatible reaction conversion rates.
• To demonstrate the utility of ML in selecting building blocks for DEL synthesis, particularly for challenging reactions.
• To enable the use of difficult reactions in DEL synthesis by pre-screening building blocks.

Main Methods:

• Development of a machine learning model to predict the conversion rate of building blocks with a model DNA-conjugate.
• Application of the ML model to assess building blocks for the Pictet-Spengler reaction, a challenging transformation.
• Validation of the ML model's predictions through experimental testing.

Main Results:

• The developed ML algorithm accurately predicts building block conversion rates in DNA-compatible reactions.
• The ML approach successfully identified suitable building blocks for the Pictet-Spengler reaction, overcoming its inherent challenges.
• This method allows for the use of reactions with historically low building block pass rates in DEL synthesis.

Conclusions:

• Machine learning can significantly enhance DNA-encoded library synthesis by predicting reaction outcomes.
• The developed ML tool facilitates the selection of building blocks, reducing experimental effort and cost.
• This approach broadens the scope of reactions amenable to DEL technology, accelerating drug discovery.