ADME Properties in Drug Discovery

Measuring ADME parameters with the highest accuracy can only be achieved in vivo, although even in this context, variability typical of a living test system may occur. Preliminary insights are often gained using animal models or other surrogate test systems. However, all of these approaches share a high cost factor, which is significantly greater when aiming for reliable assessments compared to measuring a compound's potency. Therefore, most evaluations are only conducted once there are several promising drug candidates identified during the lead optimization process.

To avoid unpleasant surprises in advanced campaigns, it has always been of utmost importance to predict ADME properties. Filtering out unattractive molecules early on can save resources and thus contribute to the campaign's success.

There are various approaches to predicting these descriptors. Given that biological processes in the human body are highly complex and involve multidimensional factors—many of which are still not fully understood today—most predictive methods are based on a fundamental principle: extrapolating properties from previously acquired knowledge.

The idea behind predicting ADME properties is relatively straightforward: Starting with a large dataset of data points that link specific parameters to the 2D structure of a compound, patterns and correlations are identified. For instance, lipophilic groups increase the logP value, hydrophilic groups raise the logS value and enhance excretion while lowering the half-life, and so on. The models generated can vary in complexity, ranging from incremental calculations based on the presence or absence of structural elements (e.g., if something is present, parameter x increases; if absent, parameter y changes) to advanced machine learning algorithms that can detect more complex relationships.

The accuracy of predictions improves with the size of the dataset and the chemical diversity of the compounds covered. Today, predicting ADME properties using appropriate models is a core component of the R&D process. In most cases, it is integrated into the automated decision-making system that determines which molecules should be synthesized. For instance, if the model predicts that a molecule has highly unfavorable parameters, it is flagged accordingly and is only approved for synthesis in exceptional cases.

Given that this approach has already resulted in significant cost savings, several pharmaceutical companies have joined forces under the Mellody project to share in-house testing data to expand their dataset.

In summary, it's never too early to use ADME properties as a legitimate filter for evaluating compounds and drug candidates. While significant fluctuations in these parameters can occur throughout the hit-to-lead process, leading pharmaceutical companies have shown that applying a pre-selection strategy can effectively conserve resources.

In collaboration with Optibrium, we feature the calculation of highly relevant ADME parameters within our two platforms. These calculations can be performed instantaneously for both molecule sets and individual generated compounds, allowing on-the-fly insights into the effects of potential modifications.
The computed parameters include the following and can be easily supplemented with custom models created or shared by the community using StarDrop:

• CYP2C9 pKi
• CYP2D6 affinity category
• blood-brain barrier classification and CNS penetration
• HIA category
• P-gp category
• PPB90 category
• hERG pIC₅₀
• logD
• logP
• logS and logS at pH 7.4