Big pharma is learning from past failures to improve its ability to predict drug toxicity earlier in the development process.
Dr Dimitri Mikhailov, US team lead in cheminformatics at the Novartis Institutes for Biomedical Research, recently outlined how the pharma giant has set about predicting the safety of drug candidates. Safety has always been a key issue in drug research but since Merck & Co. was forced to pull its painkiller Vioxx (rofecoxib) in 2004, when it was discovered the drug increases the risk of heart problems, it has never been more in the public eye.
Mikhailov, told delegates at the recent drug Discovery and Development of innovative Therapeutics (DDT) conference in Boston, US, that although there are several commercially available tools for predicting toxicity and these form a "good basis" for research, they don't include internal scientific knowledge. There are a myriad of reasons a drug could have safety problems. These include the molecule's properties (adsorption, distribution, metabolism and excretion, ADME); species gender and phenotype; disease state, lifestyle and age; drug-drug interactions; and also idiosyncratic toxicities.
In 2003, Novartis began developing its own computational programme to help predict some of these issues, called ToxCheck. Since there is little point having a tool that gives excellent data but non-expert users don't know what to do with that data, Mikhailov explained that ensuring all Novartis researchers could make use of the results was high on the developers' wish list. "Clinical safety and toxicology account for around 30 per cent of drug development failures, yet although toxicity models are often easy to use, they can be difficult to interpret."
The company set about developing a programme that could be accessed from any desktop computer, was easy to use and interpret and was fast and interactive, he continued. In that way, it could form a fundamental part of the decision making process at all levels. Using public and internal data on thousands of compounds, Novartis developed a system that correlates a compound's structure with potentially toxic properties, such as cardiotoxicity and genotoxicity. This information is then used to develop toxicity alerts based on around 160 molecule substructures that Novartis believes can cause drug safety problems.
However, it still wasn't good enough, said Mikhailov and so the company has recently begun updating the software. An index of adverse reactions and in vitro profiling data are inputted into computational models trained on the aforementioned chemical 'fingerprints'. These are then used to predict off-target binding and side-effects. For example, Novartis now has a quantitative structure-activity relationship (QSAR) model that can predict if a compound is likely to interact with hERG (named after the Ether-a-go-go gene in fruit flies), which is critical in repolarising the heart and maintaining cardiac rhythm. Blocking the effect of this protein can cause arrhythmia and lead to heart attacks.
By connecting the predictions with structural features, even if a molecule is flagged by a so called 'toxicity alert', there is a chance Novartis scientists can change that part of the molecule and fix the problem, whilst maintaining its therapeutic potential, explained Mikhailov. The final programme is now used by thousands of chemists and biologists in Novartis but Mikhailov was keen to stress that the researchers can use this information as a warning sign for experimental follow up - rather to kill a compound's development straight off.
With safety concerns recently surfacing regarding GSK's Avandia (rosiglitazone), the pharma industry is increasingly seeking to reassure the public that it is doing everything it can to ensure drugs are safe by the time they reach the market. Not only this, but it also needs to reduce the number of costly drug failures, especially in late-stage development.
One of the best ways of doing this is improving scientists' ability to foresee safety issues as early as possible. Combining the latest computer technology with scientific expertise enables this to happen at the lead optimisation stage - before the drug is tested in animals or humans.