Guilt By Association
When clever chemists come up with a new medicine there are two essential hurdles to clear. First, does it work, and second, is it safe? Various phases of clinical trials are set in motion to answer both questions. The first—efficacy—is easier because one knows what to look for, i.e., the patient gets better. The second—safety—is harder because toxic side effects are unpredictable, and subtle or rare ones aren’t always apparent until late in the drug development cycle or even years after a drug has been put into widespread use. The challenge is rendered even harder because toxicity testing is tedious, expensive, and requires intensive use of laboratory animals to track down indications and mechanisms.
The problem would become ever so much more approachable if one could predict toxicity with a computer. Collaborative work from a couple of innovative biotech outfits has taken a meaningful step towards making this dream a reality. The researchers used 656 FDA-approved drugs and 73 protein targets known to be associated with adverse drug reactions. The computer calculates whether a given drug will bind to a target based on chemical pattern matching to other drugs already shown to interact with that target (cleverly dubbed by the authors “guilt by association”). The result was an approximately 50% batting average, i.e., predicted association of drugs with targets was confirmed experimentally about half the time.
This is a good result even though there are a lot of false positives. The method does give an early clue about what toxicity might be expected from a new drug entity and allows for exploration of the ramifications before too much time and treasure is invested in developing an unpromising drug. The method is admittedly imperfect because one can’t know if there are false negatives or toxicities that are simply overlooked. It’s cheap and fast, though, so a good start on improving the difficult drug development pipeline.
Tom Tritton is President and CEO of CHF.