Too many new drug compounds fail in late-stage clinical trials due to inefficacy and safety concerns. With often-cited costs of $800 million to bring a new drug to market, a compelling need exists for improved early-stage screening. Insufficient knowledge of potential human toxicity can turn a promising lead candidate into a devastating loss. Emerging technologies that will change the existing paradigm were highlighted at Mondial Research Group’s recent conference on “Predictive Human Toxicity and ADME/Tox Studies (Absorption, Distribution, Metabolism, Elimination, and Toxicity)”.
“Every drug that fails in a clinical trial or after it reaches the market due to some adverse effect was ‘bad’ from the day it was first drawn by the chemist,” stated Robert Fraczkiewicz, Ph.D., team leader, ADMET cheminformatics at Simulations Plus. “Although state-of-the-art in silico structure-property prediction tools are not yet able to predict every possible toxicity for new molecular structures, they are able to predict many with good enough accuracy to eliminate poor molecules before synthesis.”
Just like humans with their individual characteristics such as age, weight, and height, calculated molecular descriptors (e.g., number of atoms, molecular weight, shape, and other parameters) quantitatively describe molecular structures.
Simulations Plus’ ADMET Predictor software mathematically correlates measured chemical-compound properties with their molecular descriptors to build predictive models. Currently, the application offers 133 predictive models and can process approximately 200,000 compounds per hour on a personal computer, Dr. Fraczkiewicz explained.
In silico usage is only widespread for a few select properties that have had long-time usage as primary indicators of potential new molecule druggability such as logP predictions (octanol-water partition coefficient), Dr. Fraczkiewicz noted. LogP predictions measure molecular hydrophobicity, which affects drug absorption, bioavailability, hydrophobic drug-receptor interactions, and may affect toxicity.
Current ionization, solubility, and permeability predictors range from simple equations to sophisticated, and quite accurate, artificial-neural-network ensembles such as the models implemented in ADMET Predictor, which are based on a combination of molecular and atomic quantum-level descriptors.
“Today, many in silico predictions have the same margin of error as their in vitro counterparts. We will not likely ever eliminate the need for in vitro or in vivo experiments for all molecules, but better in silico models can allow focusing the resources for these experiments where they are truly needed,” concluded Dr. Fraczkiewicz.