ADMET-like Screening Compound Library

ADMET (Absorption, Distribution, Metabolism, Excretion and Toxicity) profiling is essential for drug discovery and preclinical development. Favorable ADMET characteristics can significantly reduce the attrition rate of drug candidates by improving safety, bioavailability, and pharmacokinetic performance [1].

Designed to accelerate early-phase drug discovery, our ADMET-like Screening Compound Library offers over 448,000 structurally diverse small molecules selected for optimal pharmacokinetic and safety profiles. This compound collection consists of two main subsets:

• ADME Optimized Subset: 120,000 drug-like compounds with predicted favorable ADME characteristics, especially targeting skin permeability, gastrointestinal absorption and blood-brain barrier penetration
• CYP Filter Subset: 419,000 compounds screened using machine learning to reduce CYP3A4, CYP2D6, and CYP2C9 inhibition risk, minimizing potential drug-drug interactions.

These screening subsets combine cheminformatics filtering, machine learning predictions, and synthetic feasibility assessment to provide a high-quality starting point for medicinal chemistry and hit-to-lead programs. Natural product-like cores and fragment-based drug design principles are prioritized for the development of pharmacologically relevant, orally bioavailable leads.

The compound selection can be customized based on your requirements, cherry picking is available.

Please, contact us at orders@lifechemicals.com for any additional information and price quotations.

You can also be interested in our related products:

• Biologically Active Compound Library
• Natural Product-like Compound Libraries
• Lead-like Screening Compound Library
• Phenotypic Screening Libraries

Background information

ADMET principles are fundamental to assessing the efficacy and safety of pharmaceutical candidates and guiding their optimization in the early stages of drug discovery. Regulatory agencies such as the FDA and EMA require comprehensive ADMET assessments to ensure the safety and efficacy of drugs before market approval. Beyond pharmaceuticals, ADMET principles are also applied in environmental toxicology and chemical risk assessment, supporting the development of safer compounds across various industries.

Absorption is particularly significant, as it underlies the oral bioavailability of therapeutic agents. Efficient uptake in the gastrointestinal tract (GIT) facilitates achieving optimal systemic concentrations [2]. Of particular interest are natural products and their semi-synthetic derivatives, which often exhibit favorable ADMET properties even without structural optimization. These molecules are frequently utilized as core scaffolds for derivative synthesis or as components in fragment-based design pipelines [3]. Additionally, machine-learning–enhanced, fragment-oriented compound libraries are a strategic approach for selecting entities with optimal pharmacokinetic properties [4]. Overall, early integration of in silico prediction techniques into compound library construction helps mitigate toxicity risks and improve clinical trial success rates.

Figure 1. Schematic representation of a compound screening funnel integrating ADMET filters.

Compound Selection Process

To ensure practical applicability in drug discovery, we designed a dedicated Screening Library featuring screening compounds with predicted favorable ADMET profiles, using a comprehensive, tiered filtration approach:

• In silico ADMET filters were applied to access oral bioavailability, CNS delivery and skin absorption potential
• PAINS and structural alerts were excluded to reduce assay interference
• Machine learning prediction enabled the pre-selection of compounds with low CYP450 inhibition risk

ADMET Property Subset

To construct this Screening Library, we implemented a tiered filtration workflow that incorporates physicochemical and ADMET-relevant parameters, grouped into three major pharmacokinetic profiles, guided by targeted selection criteria:

1. Skin permeability: (QPlogKp > −2.0, PSA < 90 Ų, clogP 1.0–3.0, MW < 500 Da)
2. Oral absorption (Caco-2 permeability > 25 nm/s, human oral absorption > 80 %, PSA < 140 Ų, MW 100–500 Da, ≤ 10 rotatable bonds)
3. Blood-brain barrier (BBB) penetration (QPlogBB > 0, PSA < 90 Ų, clogP 1.0–4.0, < 1 HBD, ≤ 3 HBA, MW ≤ 450 Da)

This three-tiered filtering strategy enabled the selection of more than 120,000 compounds with optimized physicochemical properties suitable for dermal, gastrointestinal and central nervous system delivery pathways.

CYP Filter Subset

Cytochrome P450 (CYP) enzymes, particularly CYP3A4, CYP2D6, and CYP2C9, play a major role in drug metabolism. Their inhibition is a key source of adverse drug-drug interactions and metabolic liabilities. Early predicting and avoiding CYP inhibition is critical to drug development success and minimizing adverse reaction risk.

Traditional experimental methods for assessing CYP inhibition are often resource-intensive and time-consuming, making machine learning (ML) a valuable alternative. By leveraging large datasets, machine learning (ML) models can rapidly and accurately predict CYP inhibitory potential based on molecular structures and properties, enabling efficient early-stage screening of drug candidates. These predictive models streamline the drug development process, reduce costs and improve the identification of compounds with favorable safety and metabolic profiles.

This Screening Subset was constructed using machine learning models trained on ChEMBL data for CYP inhibition, specifically focusing on:

• CYP3A4 (the most abundant cytochrome P450 (CYP) enzyme in the human liver; metabolizes ~30 % of marketed drugs)
• CYP2D6 (metabolizes ~20 % of drugs, in particular antidepressants, antipsychotics and beta-blockers; exhibits high genetic variability)
• CYP2C9 (metabolizes nonsteroidal anti-inflammatory drugs (NSAIDs), anticoagulants, etc.)

A deep neural network was developed using the DeepChem Python package, with 1024 molecular descriptors as input features. The performance of the trained model was evaluated using the Receiver Operating Characteristic (ROC) curve analysis and a confusion matrix visualization (see Figure 1). Quantitative evaluation was further supported by calculating the Area Under the ROC Curve (AUROC) for each class. Finally, the proprietary was screened to predict CYP inhibitory potential of individual compounds included in this selection.

Figure 2. Distribution of hepatic cytochrome P450 isoforms roles. The figure categorizes the three primary types of CYP enzymes—CYP3A, CYP2C, and CYP2D—emphasizing their relative abundance in the human liver and their roles in drug metabolism. As general, CYP3A enzymes are the most prevalent, contributing to the metabolism of over 50 % of clinically used drugs, followed by CYP2C and CYP2D families.

Figure 3. Confusion matrices for test datasets after training for Cyp2c9 and CYP2D6 respectively.

Reference:

1. Di, L., Kerns, E. H., Fan, K., McConnell, O. J., & Carter, G. T. (2003). High throughput artificial membrane permeability assay for blood–brain barrier. Drug Discovery Today, 8(21), 955–961. https://doi.org/10.1016/j.drudis.2003.12.008.
2. Doak, B. C., Over, B., Giordanetto, F., & Kihlberg, J. (2014). Oral druggable space beyond the rule of 5: insights from drugs and clinical candidates. Bioorganic & Medicinal Chemistry, 22(9), 2746–2754. https://doi.org/10.1016/j.bmc.2011.05.029.
3. Harvey, A. L., Edrada-Ebel, R., & Quinn, R. J. (2015). The re-emergence of natural products for drug discovery in the genomics era. Natural Product Reports, 32(10), 1231–1242. https://doi.org/10.1039/C5OB01031D.
4. Chen, H., Engkvist, O., Wang, Y., Olivecrona, M., & Blaschke, T. (2018). The rise of deep learning in drug discovery. Drug Discovery Today, 23(6), 1241–1250. https://doi.org/10.1016/j.drudis.2019.11.014.
5. Gong, C. et al., 2024. Evaluation of machine learning models for cytochrome P450 3A4, 2D6, and 2C9 inhibition. Journal of Applied Toxicology, 44(7), pp.1050–1066. https://doi.org/10.1002/jat.4601.