A novel methodology on distributed representations of proteins using their interacting ligands

Area of Science:

• Bioinformatics
• Computational Biology
• Cheminformatics

Background:

• Effective protein representation is vital for numerous bioinformatics tasks.
• Proteins with similar functions often bind to similar ligands.
• Ligand chemical characteristics reflect protein properties, enabling ligand-based approaches.

Purpose of the Study:

• To propose SMILESVec, a novel method for protein representation using Simplified molecular input line entry system (SMILES) strings of ligands.
• To develop a new method for computing protein similarity based on their ligands.
• To evaluate the performance of ligand-based protein representation in clustering tasks.

Main Methods:

• Proteins are represented using word embeddings of their corresponding ligand's SMILES strings.
• Protein similarity is computed based on these ligand-derived representations.
• Performance is evaluated using protein clustering with TransClust and MCL algorithms.

Main Results:

• Ligand-based protein representation using SMILES strings performs comparably to traditional protein sequence-based methods in clustering.
• This method demonstrates the efficacy of using only ligand information for protein description.
• The approach offers a viable alternative to sequence or structure-based protein representations.

Conclusions:

• Ligand-based protein representation is a promising alternative to sequence or structure-based methods.
• SMILESVec can be applied to various bioinformatics problems, including protein-ligand interaction prediction and function annotation.
• This novel approach enhances the understanding and representation of proteins in computational biology.