Predicting the Price of Molecules Using Their Predicted Synthetic Pathways

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Machine learning

Background:

• Numerous metrics like QSAR models and docking scores aid in filtering virtual molecular libraries for synthesis.
• Existing metrics do not account for the cost of novel virtual molecules, including starting material availability and price.
• Accurate cost prediction can significantly improve and accelerate decision-making in drug development and cost-of-goods analysis.

Purpose of the Study:

• To investigate the utility of predicted retrosynthetic pathways and starting material prices as features for predicting novel molecule costs.
• To develop and present a deep learning model, RetroPriceNet, for accurate molecule price prediction.
• To integrate synthetic feasibility and starting material economics into computational drug design.

Main Methods:

• Utilized Computer Aided Synthetic Planning (CASP) to predict retrosynthetic pathways for virtual molecules.
• Developed a deep learning model, RetroPriceNet, leveraging predicted synthetic routes and starting material prices.
• Trained and evaluated the model on a dataset of molecules with known synthetic pathways and associated costs.

Main Results:

• RetroPriceNet demonstrated superior performance in predicting molecule prices compared to the state-of-the-art model on a holdout test set.
• The model effectively incorporates synthetic pathway information and starting material costs into its price predictions.
• The developed approach provides a novel metric for assessing the economic viability of synthesizing virtual compounds.

Conclusions:

• Predicted retrosynthetic pathways and starting material prices are valuable features for estimating novel molecule costs.
• RetroPriceNet offers a powerful tool for accelerating decision-making in drug discovery by providing accurate cost predictions.
• This work highlights the potential of integrating computational synthesis planning with economic factors in molecular design.