Anonymous Pattern Molecular Fingerprint and its Applications on Property Identification

Area of Science:

• Cheminformatics
• Computational Chemistry
• Machine Learning

Background:

• Molecular fingerprints are crucial for representing molecular structures in cheminformatics.
• Existing methods may lack comprehensive perception of molecular interactions and structural significance.

Purpose of the Study:

• To introduce and evaluate Anonymous-FP, a novel molecular fingerprint.
• To assess its effectiveness in molecule classification and property identification.

Main Methods:

• Generating anonymous atom chains from molecular structures.
• Embedding fingerprints into vectorial space using the PV-DBOW (Natural Language Processing) technique.
• Evaluating Anonymous-FP on molecule classification tasks across various databases.

Main Results:

• Anonymous-FP demonstrates rich information content and full structural significance.
• Achieved high experimental performance, with classification accuracy exceeding 93% on NCI datasets.
• Identified atom chain scale (typically r=8) as significant for representation ability.

Conclusions:

• Anonymous-FP offers a powerful approach for molecular representation with less prior knowledge dependence.
• The method shows significant advantages in molecule classification and property identification.
• The findings highlight the importance of atom chain scale in molecular fingerprint design.