OptADMET: A Web-Based Tool for Improving ADMET Properties

Professors Cao Dongsheng from Central South University and Hou Tingjun from Zhejiang University, along with Carbon Silicon Wisdom, National University of Defense Technology, Hong Kong Baptist University, Hunan University, and Tianjin Supercomputing Center, published a research paper titled “OptADMET: a web-based tool for substructure modifications to improve ADMET properties of lead compounds” in the authoritative journal Nature Protocols.

Optimizing lead compounds is a critical step in the drug discovery process, aimed at designing potential drug candidates from biologically active compounds. During the optimization of lead compounds, potential active compounds need to undergo careful modifications to improve their absorption, distribution, metabolism, excretion, and toxicity (ADMET) characteristics. A key question faced by medicinal chemists is which compounds should be synthesized next and how to balance various ADMET properties. Molecular transformation rules derived from a large amount of experimental analysis data are crucial for improving this decision-making process.

Based on this, the authors proposed a molecular optimization guideline based on matched molecular pairs and built a high-quality, comprehensive public chemical transformation platform (https://cadd.nscc-tj.cn/deploy/optadmet/). It provides chemical transformation rules for 32 ADMET characteristics and utilizes collected experimental data for lead compound optimization. The multi-attribute transformation rule database contains a total of 41,779 validated transformation rules generated by analyzing 177,191 reliable experimental datasets. Additionally, through QSAR modeling on 239,194 molecular data, 146,450 rules were predicted as supplementary rules. Based on this vast and reliable rules database, OptADMET can identify ideal substructure transformation rules and efficiently guide the multi-parameter optimization of any query molecule. Furthermore, to enhance decision-making performance, OptADMET provides comprehensive ADMET evaluations for all optimized molecules of the query molecule.

OptADMET aims to provide guidance for experimental scientists and computational scientists, making the optimization experience of ADMET easy. OptADMET has significant advantages compared to other existing methods because it is based on matched molecular pair analysis (MMPA) from experimental data, which provides more reliable guidance for structural modifications. Additionally, it allows simultaneous optimization of up to two ADMET properties, addressing the property balance issues caused by multi-step optimization work.

Molecular optimization is divided into four steps (Figure 1C): (1) Input molecule and conditions; (2) Generate molecular results; (3) Final screening; (4) View result details and file download.

(1) Input molecule and conditions: Molecule input supports SMILES and drawing. Sample cases can be selected. Then select the rules database and molecular properties. The number of properties to be optimized simultaneously is limited to two. For example, you can choose the “Experimental Database” and the properties of “[Toxicity] hERG” and “[Absorption] Caco-2”.

(2) Generate molecular results: After the calculation is completed, OptADMET displays all matches and related statistics. The interface shows the original molecular information, the distribution map of generated molecules, attribute interval statistics, etc., and provides options for PDF reports and detailed information.

(3) Final screening: After preliminary screening, more detailed screening can be performed. Click the “Final Screening” button on the results page. After the calculation is completed, the system will display the final screening results, including original molecular information and the list of generated molecules. The “Rule ID” can be clicked to view detailed information about the transformation rules used for each molecule.

(4) Select molecular details and download files: Users can download a PDF report containing the optimization results of all molecules or only download the PDF report for selected molecules. A CSV file can also be downloaded to view the predicted information of 32 attributes.

Figure 1 | Description of the OptADMET platform. (A) Platform design, (B) Drug optimization workflow, and (C) Schematic diagram of optimization steps.

This paper uses the Hussain and Rea algorithm to generate MMPs, which fragments molecules and stores them in an inverted file structure. This paper allows molecular variations to contain up to 12 heavy atoms and cuts with up to three bonds. To consider contextual information, a set of localized atoms is included, divided into ten types.

In this study, “rules” refer to chemical transformation rules extracted from a sufficient number of MMPs to change pharmacological or physiochemical endpoint properties. To ensure the credibility of the rules, at least 10 MMPs are required for continuous values, and at least 5 MMPs for binary classification values. Thus, we obtained 41,779 experimental transformation rules and 146,450 extended transformation rules for molecular optimization (see Figure 2). Statistical significance tests were conducted using the Wilcoxon signed-rank test and binomial distribution test, with the tested rules marked as ‘highly credible rules’, and those not passing the test but meeting the minimum MMP requirements marked as ‘qualified rules’.

Figure 2 | Distribution of transformation rules. The number of transformation rules for 32 different ADMET properties in the OptADMET database.

The database interface is divided into single property database and dual property database, which contains information about the selected transformation rules, and users can filter based on their set conditions. Clicking on the graphic of the transformation rules can view detailed information.

Figure 3 | Result interface of dual property transformation rules under the experimental database for LogD7.4 and LogP.

Next, the authors focus on optimizing the hERG toxicity of a lead compound. hERG is a potassium channel that plays an important role in cardiac cells, and inhibition of the hERG channel by drugs can lead to QT interval prolongation, potentially causing arrhythmias and cardiotoxicity. Therefore, researchers evaluate the potential cardiac toxicity risk of compounds early in drug development and carry out structural optimization of lead molecules. This case study uses the OptADMET platform by inputting the SMILES of the lead compound and selecting the ‘[Toxicity] hERG’ property, matching 45 transformation rules during the calculation process, generating over 200 molecules. Through ‘Final Screening’, five molecules with no hERG toxicity were successfully predicted, one of which introduced a carboxyl group and has been reported in the literature with a Ki value of 76 μM for hERG, indicating that this optimization successfully reduced hERG activity. Additionally, the SA score displayed on the page is 3.26, indicating that the optimized molecule has relatively high synthetic accessibility. This optimization case demonstrates the successful optimization of hERG toxicity for the guiding compound (see Figure 4).

Figure 4 | (a) Initial molecule before optimization, (b) Distribution of predicted probability values for hERG of the generated molecules, and (c) Five generated molecules without hERG toxicity.

In summary, OptADMET is a web-based comprehensive platform designed to improve the ADMET properties of potential drug candidate compounds. It provides a library of chemical transformation rules and its extended library, as well as functionalities for multi-parameter optimization of ADMET. The development of OptADMET provides medicinal chemists with a valuable resource to guide and accelerate the drug optimization process.

Yi, J., Shi, S., Fu, L. et al. OptADMET: a web-based tool for substructure modifications to improve ADMET properties of lead compounds. Nat Protoc (2024). https://doi.org/10.1038/s41596-023-00942-4