Grammar-driven SMILES standardization with TokenSMILES

Area of Science:

• Computational Chemistry
• Cheminformatics
• Linguistic Principles

Background:

• Simplified Molecular Input Line Entry System (SMILES) notation presents redundancy challenges, where multiple strings can represent the same molecule.
• This redundancy complicates computational chemistry and cheminformatics tasks, hindering efficient data processing and analysis.

Purpose of the Study:

• To introduce TokenSMILES, a novel grammatical framework designed to standardize SMILES notation.
• To mitigate SMILES redundancy by transforming strings into structured, context-free sentences.
• To enable controlled generation and manipulation of valid SMILES strings with enhanced syntactic and semantic consistency.

Main Methods:

• Developed TokenSMILES, a framework applying five syntactic constraints (e.g., branch limitations, balanced parentheses) to SMILES strings.
• Utilized semantic parsing rules to ensure valence and octet compliance.
• Implemented TokenSMILES into the open-source SmilX tool for generating standardized SMILES.

Main Results:

• TokenSMILES substantially reduces SMILES redundancy for alkanes while maintaining chemical validity.
• The SmilX tool, incorporating TokenSMILES, generates valid SMILES with accuracy comparable to existing methods for molecules with low hydrogen deficiency (HDI ≤ 4).
• The framework demonstrates applicability beyond alkanes through modifications like bond insertion, cyclization, and heteroatom substitution.

Conclusions:

• TokenSMILES formalizes SMILES syntax into a standardized, machine-interpretable form, addressing a key challenge in cheminformatics.
• The framework facilitates systematic chemical space exploration for applications in drug discovery, materials design, and machine learning.
• Further development is needed for highly unsaturated systems, highlighting the importance of dynamic feasibility checks.