Designing solvent systems in chemical processes using self-evolving solubility databases and graph neural networks
Проектирование систем растворителей в химических процессах с использованием саморазвивающихся баз данных растворимости и графовых нейронных сетей
2023-07-06
SCID: 54.1/99hytm4q
Discuss with AI
graph neural networkspartition coefficientssemi-supervised self-trainingsolubility databasessolvent selection
Figures from the paper
Abstract (AI)
Designing solvent systems is key to achieving the facile synthesis and separation of desired products from chemical processes, so many machine learning models have been developed to predict solubilities. However, breakthroughs are needed to address deficiencies in the model’s predictive accuracy and generalizability; this can be addressed by expanding and integrating experimental and computational solubility databases. To maximize predictive accuracy, these two databases should not be trained separately, and they should not be simply combined without reconciling the discrepancies from different magnitudes of errors and uncertainties. Here, we introduce self-evolving solubility databases and graph neural networks developed through semi-supervised self-training approaches. Solubilities from quantum-mechanical calculations are referred to during semi-supervised learning, but they are not directly added to the experimental database. Dataset augmentation is performed from 11,637 experimental solubilities to >900,000 data points in the integrated database, while correcting for the discrepancies between experiment and computation. Our model was successfully applied to study solvent selection in organic reactions and separation processes. The accuracy (mean absolute error around 0.2 kcal/mol for the test set) is quantitatively useful in exploring Linear Free Energy Relationships between reaction rates and solvation free energies for 11 organic reactions. Our model also accurately predicted the partition coefficients of lignin-derived monomers and drug-like molecules. We anticipate this approach will be attractive to other areas of predictive chemistry where experimental, computational, and any other heterogeneous data sources should be combined.
Key Findings
1
Quantum-mechanical solubilities guide semi-supervised learning but are not directly incorporated into the experimental database, addressing differing error magnitudes and uncertainties.
2
The approach expanded 11,637 experimental solubility measurements to more than 900,000 integrated data points while correcting discrepancies between experimental and quantum-mechanical results.
3
The model accurately predicted partition coefficients for lignin-derived monomers and drug-like molecules and supported solvent selection in synthesis and separation processes.
4
The model achieved a test-set mean absolute error of approximately 0.2 kcal/mol, enabling quantitatively useful Linear Free Energy Relationship analysis for reaction rates and solvation free energies across 11 organic reactions.
5
The study introduces self-evolving solubility databases and graph neural networks using semi-supervised self-training to integrate experimental and computational information.
Research Object
Solvent systems and solubility data for chemical processes, including organic reactions, separation processes, lignin-derived monomers, and drug-like molecules
Research Subject
Solubility prediction and solvent-selection performance, including the reconciliation of experimental and quantum-mechanical data and prediction of solvation free energies and partition coefficients
Publication Details
Publication Date
2023-07-06
Journal
Publisher
ISSN
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest