MUBD‐DecoyMaker 2.0: A Python GUI Application to Generate Maximal Unbiased Benchmarking Data Sets for Virtual Drug Screening

Abstract

Ligand enrichment assessment based on benchmarking data sets has become a necessity for the rational selection of the best‐suited approach for prospective data mining of drug‐like molecules. Up to now, a variety of benchmarking data sets had been generated and frequently used. Among them, MUBD‐HDACs from our prior research efforts was regarded as one of five state‐of‐the‐art benchmarks in 2017 by Frontiers in Pharmacology. This benchmarking set was generated by one of our unique de‐biasing algorithms. It also rendered quite a few other cases of successful applications in recent years, thus is expected to have more impact in modern drug discovery. To make our algorithm amenable to more users, we developed a Python GUI application called MUBD‐DecoyMaker 2.0. Moreover, it has two new additional functional modules, i. e. “Detect 2D Bias” and “Quality Control”. This new GUI version had been proved to be easy to use while generate benchmarking data sets of the same quality. MUBD‐DecoyMaker 2.0 is freely available at https://github.com/jwxia2014/MUBD‐DecoyMaker2.0, along with its manual and testcase.

AutoMolDesigner for Antibiotic Discovery: An AI-Based Open-Source Software for Automated Design of Small-Molecule Antibiotics

“…coli, 4774 known antibiotics with “high activity” (see Data Curation for details) were used for modeling. They were used as input to make benchmarking sets by MUBD-DecoyMaker2.0 . The benchmark for model training was composed of 501 positive data points (unbiased ligands) and 19,539 negative data points (unbiased decoys), which simulated the VS scenario in the real world (i.e., a low hit rate in HTS, Table S5).…”

AutoMolDesigner for Antibiotic Discovery: An AI-based Open-source Software for Automated Design of Small-molecule Antibiotics

“…were used for modeling. They were firstly used to make benchmarking sets by MUBD-DecoyMaker2.0 37 , a publicly accessible GUI application focused on making trustworthy datasets for virtual screening. The benchmark for model training was comprised of 501 positive data points (unbiased ligands) and 19,539 negative data points (unbiased decoys), which simulated the virtual screening scenario in real world (i.e., a low hit rate in HTS, Table S4).…”

“…In the section of “Tool II: AutoML-based Molecular Property Prediction for Small-molecule Antibiotics”, MUBD-DecoyMaker2.0 7 was used to make the dataset used for training and predicting antibacterial activity against E. coli. .…”

A SAR and QSAR study on cyclin dependent kinase 4 inhibitors using machine learning methods

“…Training set 2 contains 2263 inhibitors, 1231 of which are highly active, and based on these highly active inhibitors, we generated 6085 decoys, training set 2 and these decoys form training set 4, which includes 8348 molecules. We used MUBD-DecoyMaker2.0 46 to generate decoys based on the test set. Test set 1 contains 752 inhibitors, 438 of which are highly active, and based on these highly active inhibitors, we generated 2377 decoys, test set 1 and these decoys form test set 3, which includes 3129 molecules.…”