Three Dimensional Approaches to Predicting Toxicity

3D approaches use knowledge about the shape and volume of either the submitted compound or the binding site thought to be involved in the mechanism of toxicity, or both. They often require knowledge that the chemicals act via a common mechanism or biological event. One example where 3D methods have been applied to predict toxicity is the binding or blockage of the potassium ion channel in the prediction of prolonged QT syndrome, another is the identification of substrates for one or more of the P450 cytochromes and their ability to cause induction or peroxisome proliferation. P450 induction where a patient is taking multiple therapeutic medicines has been linked to causing adverse events through prolonged or increased exposure to one of the drugs or decreased efficacy by increased metabolic clearance. This type of effect is often referred to as a drug-drug interaction (DDI).

As with the prediction of efficacy using 3D models, development of a model for toxicity requires, at a minimum, a set of chemicals with a range of known activities for a particular endpoint. If one assumes that these compounds all act via common interactions with proteins it is possible to develop a pharmacophore that describes the essential arrangement of structural components for a compound to express similar biological activity. Other techniques such as volume or molecular planarity estimates can also be used where there is little knowledge of the protein or its binding site.

Similarly, techniques such as comparative molecular field analysis (CoMFA),20'21 developed by Tripos Inc., can also be applied in these types of situations. CoMFA models use estimates of steric and electrostatic interactions to describe the activities of molecules. The primary limitation in the development of CoMFA models is the need for good structural alignment between not only the molecules used to build the model but also those molecules for which predictions are made.

It is often preferable, although not necessary, to know which protein is involved in the expression of toxicity and its active binding site. This additional knowledge enables the developer to include more accurate estimates of volume restrictions and alternative conformations that may further enhance the prediction of protein-ligand interactions. Techniques such as ligand docking and scoring can be used to rank order chemicals according to their ability to bind to the receptor site. These techniques often have to compensate for flexibility not only in the ligand being docked but also the protein-binding site itself. This can be accomplished by either using multiple conformations of the compound or using energy minimization algorithms after the docking has been performed.

As with statistical methods, 3D models can suffer from over-fitting, which results in the model being too specific to the training set of compounds, reducing its predictive power. It is therefore common for developers to use separate training and test sets to ensure that over-fitting does not occur. Advantages of three-dimensional approaches

• Predictions from these 3D approaches are usually based on mechanistic hypotheses of toxicity, and therefore usually provide reliable predictions where the molecule in question acts via this mechanism and is adequately represented in the model training set.

• These approaches take a whole-molecule view to predicting their activity, and can compensate for subtle variations in electronic or steric factors within the structure.

• The predictions made by 3D methods are often numerical estimates of the activity of a molecule, and therefore these approaches are capable of providing a rank ordering of novel chemical structures. Disadvantages of three-dimensional approaches

• They rely on the assumption that the compounds in question cause toxicity through their interaction with specific proteins or receptors. Toxicity can also occur as a result of direct covalent binding to proteins or as a result of altering the physical state of the biological system for example by increasing or decreasing intracellular pH levels.

• 3D methods are often computationally intensive and require extensive calculation times to generate predictions. While this can be compensated for by parallelization, it ultimately leads to some level of approximation in the calculation, which can lead to errors in the predictions from the models.

• Often it is difficult to reliably confirm that all the molecules in the training set are acting through a common mechanism or even through interactions at the same binding site within the protein complex. These assumptions can lead to errors in the predictions from these models.

• Most types of toxicity studies do not yield data that are amenable to this type of model development, as often experiments are conducted with a view to observing changes at an organism level and not at the molecular level.

Due to the nature of toxicology studies there are few endpoints where specific proteins have been identified as the primary cause of the adverse effect or where a molecular shape has demonstrated good correlations with the data generated. This has naturally limited the application of 3D approaches to yield truly predictive models for toxicity. As knowledge around the mechanisms of toxicity expand and new technology facilitates more precise measurements of the interactions between molecules and a biological system, these types of approaches will most likely yield the best predictive models. However, the scope of each model needs to be adequately assessed and applied appropriately.

Lower Your Cholesterol In Just 33 Days

Lower Your Cholesterol In Just 33 Days

Discover secrets, myths, truths, lies and strategies for dealing effectively with cholesterol, now and forever! Uncover techniques, remedies and alternative for lowering your cholesterol quickly and significantly in just ONE MONTH! Find insights into the screenings, meanings and numbers involved in lowering cholesterol and the implications, consideration it has for your lifestyle and future!

Get My Free Ebook


  • Ralf
    How to predict toxicity of new drug computationally?
    1 month ago

Post a comment