Building Machine Learning Small Molecule Melting Points and Solubility Models Using CCDC Melting Points Dataset.

Area of Science:

• Computational Chemistry
• Drug Discovery
• Materials Science

Background:

• Accurate prediction of small molecule solubility is crucial but hindered by limited experimental data.
• Solubility depends on lattice dissociation energy (related to melting point) and solvation efficiency (log P).

Purpose of the Study:

• To develop widely applicable machine learning models for predicting small molecule melting points.
• To predict aqueous thermodynamic solubilities using established solubility equations and the developed melting point models.

Main Methods:

• Utilized the Cambridge Crystallographic Data Centre's (CCDC) melting point dataset of nearly 100,000 compounds.
• Developed machine learning models to predict melting points.
• Applied the general solubility equation to predict aqueous solubilities based on predicted melting points.

Main Results:

• Successfully created machine learning models for predicting small molecule melting points.
• Demonstrated the ability to predict aqueous thermodynamic solubilities using these models and the general solubility equation.
• The global model showed potential for localization with additional specific experimental data.

Conclusions:

• Machine learning models based on melting point data offer a viable approach for predicting small molecule solubility.
• This method addresses the challenge of limited experimental solubility data.
• The models can be refined for specific chemical series by incorporating additional melting point measurements.