Report comment

Not long ago, I had a discussion with a former reporter at C&EN about the phrase "drug design". Now, let's agree that if drugs could be designed the structure of the target would have to be known. Now, agree with me that having the non-in silico structure of a target is definitely insufficient for predicting a binder... with the exception of oligonucleotides. So, in my opinion we do not design drugs. The structure of a target that includes a bound molecule allows us to optimize hits; that is not 'design'. The reporter disagreed with my premise because he was able to find 15,000 hits for the phrase 'drug design' on Google. That's a bad method for determining 'truth' on so many levels.

Well, if we can't design a drug given the actual structure of a target it ain't gonna be better with the 'maybe' structure of a calculated target. Most of AlphaFold's 'structures' are just pictures, pictures generated by an AI algorithm that predicted correctly some of the time. To know which structures were predicted correctly means that a real structure existed, and AI algorithms are fed basis sets to train on. Think there might be a correlation? I'd bet my hat on it.

My friends trained in science, don't lose your skepticism. We need to be the purveyors of eternal truths, not ephemeral truths.