Comparing a Query Compound with Drug Target Classes Using 3D-Chemical Similarity

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• Three-dimensional (3D) similarity is crucial for predicting properties of novel molecular structures.
• Predicting drug targets for new compounds often relies on comparing them to known compounds, but this is challenging for structurally dissimilar molecules.
• Existing 3D similarity methods depend on factors like conformational sampling, alignment, descriptors, and similarity coefficients.

Purpose of the Study:

• To develop a quantitative method for comparing query compounds to target classes using 3D similarity distributions.
• To assess the discriminative power of 3D similarity-based metrics for predicting drug targets of unprecedented molecular scaffolds.
• To introduce Kullback-Leibler (K-L) divergence as a tool for evaluating compound-target associations.

Main Methods:

• Utilized maximum likelihood (ML) estimation to transform Jaccard-Tanimoto similarity of query-ligand pairs into query distributions.
• Employed Gaussian mixture models (GMM), optimized via the expectation-maximization (EM) algorithm, to represent target class distributions.
• Calculated Kullback-Leibler (K-L) divergence to quantify the discriminativeness of query compounds against target classes.

Main Results:

• The 3D similarity-based K-L divergence, along with probability and feasibility index (Fm), demonstrated significant discriminative power for certain query-class associations.
• The method effectively quantifies the distinction between a query compound and various target classes.
• K-L divergence proved useful for ranking 3D similarity scores and assessing the statistical significance (p-value) of similarity distributions.

Conclusions:

• Three-dimensional similarity, quantified by K-L divergence of similarity distributions, offers a valuable approach for predicting targets of novel drug scaffolds.
• This method enhances the prediction of targets for compounds that are structurally distinct from known agents.
• The K-L divergence serves as an additional metric to improve the accuracy and reliability of drug target prediction in cheminformatics.