Ranking Molecules with Vanishing Kernels and a Single Parameter: Active Applicability Domain Included

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• Ligand-based virtual screening utilizes high-throughput screening (HTS) data to train classification models.
• These models prioritize untested molecules based on predicted activity against a target protein.
• Existing methods can be computationally intensive and lack robust applicability domain definitions.

Purpose of the Study:

• To propose a novel, computationally efficient single-parameter ranking method with an Applicability Domain (AD) for ligand-based virtual screening.
• To improve Kernel Density Estimates (KDE) by incorporating vanishing kernels and Tanimoto distance for molecular fingerprints.
• To introduce the Vanishing Ranking Kernels (VRK) method.

Main Methods:

• Revisiting Kernel Density Estimates (KDE) with two modifications: using vanishing kernels and Tanimoto distance between molecular fingerprints as a radial basis function.
• Developing the Vanishing Ranking Kernels (VRK) method, a single-parameter ranking approach.
• Evaluating VRK performance on 21 high-throughput screening (HTS) assays.

Main Results:

• Vanishing Ranking Kernels (VRK) demonstrate competitive performance compared to graph convolutional deep neural networks.
• VRK models are conceptually simple and exhibit fast training times, requiring optimization of only a single parameter.
• Trained VRK models inherently define an active applicability domain (AD), enhancing screening efficiency.

Conclusions:

• VRK offers an efficient and effective alternative for ligand-based virtual screening and molecular prioritization.
• The incorporation of an applicability domain in VRK significantly improves the practical utility and screening frequency.
• The method's simplicity and speed make it a valuable tool in drug discovery pipelines.