LINGO-DL: a text-based approach for molecular similarity searching

Area of Science:

• Cheminformatics
• Computational Chemistry
• Drug Discovery

Background:

• Line notations like Simplified Molecular Input Line System (SMILES) offer compact storage and transfer of molecular structures.
• The existing LINGO method uses fixed-length substrings (LINGOs) from SMILES for molecular similarity and property prediction.
• There is a need for improved methods to capture molecular similarity, especially for complex datasets.

Purpose of the Study:

• To introduce LINGO-DL, an alternative LINGO method utilizing variable-length substrings from SMILES.
• To evaluate the performance of LINGO-DL against the original LINGO method in virtual screening tasks.

Main Methods:

• Canonical SMILES strings are fragmented into substrings of three different lengths.
• The LINGO-DL method compares these variable-length substrings to measure molecular similarity.
• Retrospective virtual screening was performed using MDDR, DUD, and MUV datasets.

Main Results:

• LINGO-DL demonstrates superior performance compared to the LINGO method.
• The improvement is most significant when screening for active molecules with high structural heterogeneity.
• This suggests LINGO-DL better captures nuanced structural relationships.

Conclusions:

• LINGO-DL offers a more effective approach to molecular similarity assessment than the original LINGO method.
• The method's ability to handle structural diversity makes it valuable for drug discovery and virtual screening.
• Variable-length fragmentation of SMILES is a promising strategy for cheminformatics applications.