Molecular Descriptors, Structure Generation, and Inverse QSAR/QSPR Based on SELFIES

Area of Science:

• Computational Chemistry
• Cheminformatics
• Drug Discovery

Background:

• Traditional inverse Quantitative Structure-Activity Relationship (QSAR) and Quantitative Structure-Property Relationship (QSPR) methods require generating numerous chemical structures and calculating their descriptors, often lacking a direct structural-descriptor correspondence.
• This limitation hinders efficient molecular design and property prediction.

Purpose of the Study:

• To propose a novel approach for inverse QSAR/QSPR using Self-Referencing Embedded Strings (SELFIES), a robust molecular representation.
• To establish a one-to-one mapping between molecular descriptors and chemical structures for improved inverse QSAR/QSPR.
• To demonstrate the successful generation of molecules with targeted properties.

Main Methods:

• SELFIES strings were converted into one-hot vectors to derive SELFIES descriptors (x).
• An inverse analysis of QSAR/QSPR models (y = f(x)) was performed using these descriptors and objective variables (y).
• SELFIES-based structure generation was employed to create molecules corresponding to specific descriptor values.

Main Results:

• The proposed SELFIES descriptors and structure generation method were validated on real compound datasets.
• SELFIES-descriptor-based QSAR/QSPR models demonstrated predictive performance comparable to existing fingerprint-based models.
• A significant number of molecules with a one-to-one relationship to SELFIES descriptor values were successfully generated.

Conclusions:

• The SELFIES representation provides a robust and efficient framework for inverse QSAR/QSPR.
• This method enables the direct generation of molecules with desired properties, overcoming limitations of conventional approaches.
• The study successfully demonstrates the application of SELFIES in generating molecules with specific target properties.