What's the right bar for computation in drug discovery? People say that the question "Has computation ever discovered a drug?" is fundamentally unfair, and I agree because it would be fundamentally unfair for any single technology. Drug discovery is a confluence of methods. 1/n
Nonetheless, in one sense the bar is simple: get computation to a stage where you show that it's a requirement rather than just a support discipline for a company's success. Have it front and center in molecular design rather than just a peripheral contributor.
Currently you cannot imagine a drug discovery company without synthetic chemists, biochemists or pharmacologists, but there are many successful drug companies without computational chemists. We would love to see that change.
For this two things need to happen, on opposite sides: other scientists need to provide computation with the support it needs to rapidly validate and discard hypotheses. But computation needs to accurately communicate what can and can't be done, and that's where hype truly hurts.
Some people say hype is fine because the more ideas are funded the merrier, and let the chips fall where they may. But beyond a certain extent this attitude is very detrimental because repeated promise-reality gaps lead to deep skepticism and even promising approaches discarded.
The problem is that when you say computation can help with everything - affinity, permeability, solubility, metabolism - it is wrong at worst and confusing at best, because there's huge variance in success for each of these. Communicating that variance is key for credibility.
Another problem is wrong institutional incentives. Computational scientists who might already think of themselves as peripheral value their jobs and are very reluctant to say computation *can't* help solve a problem. They should not be penalized for saying no.
It's a problem I've encountered: somebody would ask me to predict the permeability of a complex molecule. I know that the models are crappy, but nobody's compensating me for saying "Sorry, I can't help you. You are better off actually measuring the permeability of this compound."
An ideal organizational structure is where computational scientists are rewarded as much for saying what won't work as what would. Ultimately that would ensure that computation is applied optimally and would actually be in the company's best self-interests. Culture shift needed.
Now this raises the questions: What about validation? Unless you try it, how do you know it works? Two answers: validation needs to ideally be built into the *development* of the models, not their *application*. And yes, experimentalists need to occasionally help do this.
Bottom line IMO: Until comp chemists are more upfront about their domains of applicability, institutional incentives are set up to encourage them from saying what won't work and experimentalists are readily willing to validate models, computation will remain a support discipline.
