How do I know if Optibrium’s predictive models work?
Data curation for model building A model can only be as good as the data it has been trained on.…
The first of the ADMET properties relate to absorption. Understanding how a drug is absorbed and cleared in tissues is important in predicting the drug’s efficacy and safety. Drugs can be absorbed in the human system through passive permeation or passive absorption. Some drugs require carrier-mediated transport or transporter-mediated absorption to be absorbed.
Lots of factors influence absorption, so there are many properties to consider, such as lipophilicity (logP and logD), pKa and solubility. For example, with lipophilicity, if a drug is too hydrophilic, it will be unable to cross cell membranes to reach the final target. If a compound is too lipophilic, it may get stuck in the membrane or fatty tissues and not reach the target. So, we need that ideal balance.
When examining transporter-mediated absorption, we typically use cell-based assays, where the protein transporter is ‘ectopically expressed’ (i.e., expressed where it wouldn’t usually be) in cell lines. This enables us to study drug efflux characteristics or cell permeation. Common assay systems include: MDCK-MDR1, Caco-2, and MATE2-K assay systems.
Alternatively, we can use non-cell-based assays, and look at isolated membrane fractions that express transporter proteins. Such assays include: P-gp, BCRP, OATP1B1/1B3, OAT1/3, OCT1, and OCT2 (Tweedie et al. 2013).
StarDrop contains models for P-gp transportation, to identify potential P-glycoprotein ligands. Compounds which inhibit P-gp can block drug uptake, increasing intracellular concentrations and restoring efficacy.
It’s important to note, that although in vitro and in vivo methods have been used to estimate absorption properties, these can be costly, resource-intensive and difficult to interpret. Computational methods can overcome such hurdles. StarDrop, for example, contains a Human Intestinal Absorption (HIA) classification model, to identify compounds with good absorption based on a set of meaningful descriptors. We’ll be going into HIA in a little more detail in a separate future blog post if you’d like to know more on how and why we can predict HIA.
The distribution of a drug throughout the body depends on many factors. these include blood flow, plasma protein binding, lipid solubility, the blood-brain barrier, and the placental barrier.
Plasma protein binding (PPB) refers to the reversible binding of a drug to proteins found in blood plasma. The degree of binding to plasma proteins affects both the pharmacokinetic and pharmacodynamic properties of a drug. Specifically, only the fraction/proportion of the unbound drug in the plasma is available to exhibit pharmacologic effects and to be excreted. When a drug binds extensively to the plasma proteins, a higher amount must be absorbed to reach effective therapeutic levels. Additionally, since the binding is reversible, the bound drug can also serve as a reservoir. From this, the drug is released to maintain equilibrium as the unbound fraction is metabolised or excreted. Albumin and globulin are the most common plasma proteins that bind to drug molecules.
StarDrop’s ADME QSAR module provides a binary (low and high binding) categorical model for estimating PPB.
The blood-brain barrier is a semipermeable membrane formed by endothelial cells. It prevents the entry of harmful substances found in the blood, thereby protecting the brain.
Passing the blood-brain barrier is a challenge particularly when developing central nervous system (CNS)-targeted therapeutics and diagnostic agents. For non-CNS targets, it is still important to consider. Here, blood-brain barrier penetration can cause unwanted side effects.
StarDrop has two BBB models: a continuous model, which provides an estimate of log([brain]/[blood]), and a classification model. Since the two models have been developed independently, a consensus score provides the best approach for assessing the compound’s ability to penetrate the blood-brain barrier.
Optimising metabolism is important. Your compound needs to be stable enough to reach the target in desired concentrations, but then able to be broken down to be excreted. You also need to know which enzymes metabolise your compound to minimise the risks of drug-drug interactions or issues due to genetic polymorphisms. And going into the T of ADMET, you need to know which metabolites will form to avoid reactive or toxic metabolites and unwanted side effects.
Several enzyme families can be involved in the metabolism of a drug compound, which typically occurs in multiple phases.
Phase I metabolism includes oxidation, reduction, and hydrolysis reactions. These reactions convert lipophilic drugs into more hydrophilic molecules by adding polar functional groups or making existing ones more polar.
The most common enzyme family involved in Phase I metabolism is cytochrome P450 (CYP450). This superfamily of enzymes is responsible for the metabolism of 75 to 90% of human hepatically cleared drugs (Dixit et al., 2017). Other enzyme families involved in Phase I metabolism include flavin-containing monooxygenase (FMOs), aldehyde oxidase (AOXs), and alcohol dehydrogenase (ADHs).
During Phase II metabolism, charged (polar) groups, such as glucuronic acid, sulfate and glutathione, are added to the molecules through conjugation reactions. These reactions tend to produce more polar metabolites with an increased molecular weight. This helps reduce the drug’s pharmacological activity and ease excretion.
Enzyme families that are involved in Phase II metabolism include UDP-glucuronosyltransferase (UGT), sulfotransferase (SULT) and glutathione S-transferase (GST).
Both the Metabolism module in StarDrop and Semeta can help you generate metabolism predictions for many of these key enzyme families. They enable you to understand your compounds’ metabolic routes, products, pathways and labilities.
If you’d like to learn more about the importance and challenges of predicting drug metabolism, you can read our ebook.
A key factor in excretion, in which the body gets rid of waste products, is molecular weight. Substances of small molecular weight are mainly removed through the urine. In addition, we can predict passive excretion based on a variety of approaches. These include the flow rate, logP or lipophilicity, protein binding, and the pKa value; all of these factors may affect how drugs are reabsorbed and excreted. Otherwise, hepatic metabolism and active drug transport by biliary transport are both important to consider.
To ensure a drug’s safety, many different types of toxicity must be considered. However, modelling toxicity can be challenging. In silico models are often built solely on publicly available data. This may mean that they are less predictive for proprietary chemical space. Data-sharing initiatives can improve the performance of such models. However, organisations are often unable to share their data due to the need to protect the confidentiality of their research and development programmes.
In StarDrop’s Derek Nexus module, in silico models use expert knowledge to develop structural alerts based on chemical toxicity. These models include the usage of proprietary data to identify new areas of chemical space to build and refine existing alerts while still preserving the privacy of the confidential data. Models in Derek Nexus include endpoints for mutagenicity, skin sensitisation, and skin irritation and corrosion.
StarDrop also contains a model for hERG inhibition. This acts as a biomarker for cardiotoxicity. Inhibition of the human Ether-a-go-go-Related Gene (hERG) potassium channel appears to be the most common mechanism of acquired QT interval prolongation. QT interval prolongation has been linked to life-threatening heart conditions, such as Toursade de Pointes. As more non-antiarrhythmic drugs are shown to have the potential to prolong QT interval, it is important that all new chemical entities (NCE) are investigated for hERG inhibition early in their preclinical development. Recent reviews of predictive in silico models for hERG inhibition were given by Gola et al. and by Song and Clark (Gola, Obrezanova, Champness, & Segall, 2006).
The StarDrop ADME QSAR module offers a collection of high-quality predictive ADME models, allowing you to explore diverse chemistry and make informed choices for your synthetic strategy with confidence. Models for simple properties are also available in StarDrop as well as additional custom models that can be downloaded from the customer hub area of our website. If models tailored to specific chemistry or data are required, the Auto-Modeller module offers a convenient option to build robust QSAR models to complement the StarDrop models.
Any StarDrop models and models created using the Auto-Modeller module return an uncertainty value along with the property value. The uncertainty provides an indication of confidence in the prediction and reflects the molecule’s proximity to the chemical space of the model.
Data curation for model building A model can only be as good as the data it has been trained on.…
What’s the purpose of a predictive model? What’s the value of predictive models for drug discovery? Most of the undergraduate…
The role of generative chemistry in drug discovery A key difficulty in finding new drugs is the sheer size of…
A further advantage of using the StarDrop models, those created using Auto-Modeller or scores, is the ability to generate a Glowing Molecule depiction. The Glowing Molecule highlights the parts of the molecule having the greatest influence on the prediction. It can be helpful in guiding the redesign of a compound.
So, we’ve talked a bit about some of the key ADMET properties to consider for drug discovery. And, a bit about how to predict these. There are many more properties which we haven’t gone into in this blog post. So, with so many properties to consider (alongside activity!), how can we identify the best compounds for our objectives?
Star Drop’s Probabilistic Scoring approach to multi-parameter optimisation (MPO) allows you to assess the likelihood of success of your compounds against a project criterion by simultaneously considering properties (experimental or predicted) according to their desired value and relative importance. The resulting score (a value between 0 and 1) provides an estimate of the compound’s chance of success. It uniquely takes into account the uncertainty associated with each property, experimental or statistical.
There isn’t a one-approach fits-all solution when it comes to the ADMET properties you must consider. Depending on drug target, delivery, recipient and more, the key factors will likely be unique to a specific discovery program or project. But we hope we’ve been able to give you a good overview of a few of the ADMET properties you might want to keep in mind.
Alessia is a Senior Application Support Scientist at Optibrium, where she leverages her expertise to provide advanced technical support and guidance for drug discovery software solutions. With a PhD in Chemical Engineering from the University of Strathclyde and prior research experience at the Max Planck Institute for Polymer Research, Alessia has a strong background in chemical engineering and materials science. Her career also includes industry experience at Baxter International, where she gained practical insights into pharmaceutical processes.
Barry, our Application Support Scientist, has over 15 years experience in academia and industry. His PhD in Molecular and Cellular Biology from the University of Dundee focused on the mechanisms of cell migration, metabolism and signal transduction. He enjoys problem solving for drug discovery, bioscience and healthcare professionals, and showing users new tricks to make the most of their StarDrop software.
Summary In this article, ‘Addressing toxicity risk when designing and selecting compounds in early drug discovery‘, we discuss the application…
Develop advanced MPO strategies and target the right compounds, faster.
We’re diving back into our favourite subject: multi-parameter optimisation.
Introduction Predicting sites of metabolism (SoM) enable chemists to be more efficient in optimising the structure of new chemical entities…