Representation of Molecules by Sequences of Instructions

Area of Science:

• Computational Chemistry
• Cheminformatics
• Artificial Intelligence

Background:

• Molecular graph complexity challenges computational intelligence methods.
• Current representations like SMILES have limitations, including invalid strings and sensitivity to minor changes.
• Existing methods struggle with accurate and robust molecular representation for AI.

Purpose of the Study:

• To develop a new chemical nomenclature system for improved molecular representation.
• To address the limitations of existing standards like SMILES.
• To create a representation suitable for advanced computational intelligence and deep learning models.

Main Methods:

• Defined a reduced instruction set for chemical nomenclature.
• Developed a language based on sequences of these instructions.
• Ensured instructions allow incremental atom addition and bond modification.

Main Results:

• All strings generated by the new language represent valid molecules.
• Minor string modifications correspond to small molecular changes, enhancing robustness.
• The new representation is suitable for computational intelligence systems.

Conclusions:

• The proposed instruction-based language offers a robust and valid molecular representation.
• This approach overcomes SMILES limitations for computational chemistry applications.
• It enables more effective use of deep learning in chemical information processing.