Mol2Context-vec: learning molecular representation from context awareness for drug discovery.

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Computer-aided drug design (CADD) is crucial for drug discovery, facing challenges in accurate molecular representation.
• Current methods struggle with substructure ambiguity and information flow between atomic groups.

Purpose of the Study:

• To develop a novel deep learning architecture, Mol2Context-vec, for enhanced molecular representation.
• To address limitations in current molecular expression techniques within CADD.

Main Methods:

• Proposed a deep contextualized Bi-LSTM architecture named Mol2Context-vec.
• Integrated different levels of internal states for dynamic molecular substructure representations.
• Enabled capture of interactions between any atomic groups, including topologically distant ones.

Main Results:

• Mol2Context-vec achieved state-of-the-art performance on multiple benchmark datasets.
• The model's visual interpretations align closely with human chemical intuition.
• Demonstrated effectiveness in capturing complex atomic interactions.

Conclusions:

• Mol2Context-vec offers a reliable and effective tool for molecular expression in CADD.
• The model enhances the understanding of molecular structures and interactions.
• This approach holds significant potential for advancing drug discovery processes.