Knowledge-based model of hydrogen-bonding propensity in organic crystals

Area of Science:

• Crystallography
• Computational Chemistry
• Drug Discovery

Background:

• Predicting hydrogen bonding is crucial for understanding crystal structures and stability.
• Existing polymorph prediction methods face computational challenges.
• Identifying hydrogen bond donors and acceptors aids in rationalizing crystalline forms.

Purpose of the Study:

• To present a new knowledge-based method, the logit hydrogen-bonding propensity (LHP) model, for predicting hydrogen bond formation in crystal structures.
• To provide a tool that can identify both common and uncommon hydrogen bonding interactions.
• To support drug development by aiding in the rationalization of stable and metastable crystalline forms.

Main Methods:

• Statistical analysis of hydrogen bonds in the Cambridge Structural Database (CSD).
• Development of the logit hydrogen-bonding propensity (LHP) model using logistic regression.
• Training data derived from a novel survey method categorizing hydrogen bonds and extracting parameters from 3D organic crystal structures.

Main Results:

• The LHP model accurately predicts hydrogen bond formation using only 2D observables.
• Achieved approximately 90% correct classification for observed hydrogen bonds and non-interacting pairs in initial analyses.
• Extensive statistical validation confirmed the model's robustness across diverse small-molecule organic crystal structures.

Conclusions:

• The LHP model offers a computationally feasible, knowledge-based alternative for predicting hydrogen bonding in crystals.
• This method can serve as a valuable precursor to polymorph screening, enhancing drug development processes.
• The model's high accuracy and robustness demonstrate its potential utility in crystallographic research and pharmaceutical applications.