fragSMILES as a chemical string notation for advanced fragment and chirality representation

Area of Science:

• Computational chemistry
• Drug discovery
• Cheminformatics

Background:

• Generative models are crucial for de novo drug design, producing molecules with specific properties.
• String-based representations like SMILES and SELFIES are widely used but have limitations in capturing chirality and synthetic accessibility.
• Existing methods struggle to efficiently represent complex molecular information.

Purpose of the Study:

• Introduce fragSMILES, a novel fragment-based string representation for molecules.
• Address limitations of current string notations in de novo drug design.
• Improve the generation of molecules with desired biochemical and scaffold properties.

Main Methods:

• Developed fragSMILES using a graph-reduction approach for chemically meaningful fragment encoding.
• Ensured efficient, interpretable, and non-redundant molecular representation.
• Embedded molecular chirality information directly into the fragSMILES notation.

Main Results:

• fragSMILES effectively encodes fragments and their breaking bonds independently.
• The representation successfully embeds molecular chirality information.
• fragSMILES demonstrated promise in generating molecules with desirable biochemical and scaffold properties compared to SMILES, SELFIES, and t-SMILES.

Conclusions:

• fragSMILES offers an efficient and interpretable molecular representation for generative drug design.
• This novel notation overcomes key limitations of existing string-based methods, particularly regarding chirality.
• fragSMILES shows significant potential for advancing de novo drug design and molecule generation.