Ligand-Based Virtual Screening Based on the Graph Edit Distance

Area of Science:

• Computational Chemistry
• Cheminformatics
• Bioinformatics

Background:

• Chemical compounds are modeled as attributed graphs, with nodes representing components and edges representing chemical bonds.
• Graph Edit Distance (GED) calculates molecular dissimilarity by measuring the cost of transforming one graph into another.
• Accurate bioactivity dissimilarity requires carefully tuned transformation costs within the GED framework.

Purpose of the Study:

• To analyze structural-based screening methods for evaluating Harper's transformation cost proposal.
• To develop an algorithm for learning transformation costs to accurately define bioactivity dissimilarity.
• To optimize ligand-based virtual screening applications using learned graph edit distances.

Main Methods:

• Utilized attributed graphs to represent chemical compounds and their relationships.
• Employed Graph Edit Distance (GED) to compute bioactivity dissimilarity between molecules.
• Developed and validated a novel algorithm for learning GED transformation costs.
• Evaluated performance using six public datasets: CAPST, DUD-E, GLL&GDD, NRLiSt-BDB, MUV, and ULS-UDS.

Main Results:

• The proposed algorithm successfully learned transformation costs that accurately define bioactivity dissimilarity.
• Classification accuracy, used as a measure of dissimilarity quality, was significantly improved.
• The learned costs demonstrated superior performance in identifying bioactivity similarity within structurally diverse molecular sets.

Conclusions:

• The developed algorithm provides an effective method for learning graph edit distance costs for bioactivity prediction.
• This approach enhances the accuracy of ligand-based virtual screening by improving molecular dissimilarity measures.
• The methodology shows promise for identifying bioactive molecules in large chemical libraries.