Time to converge

The adoption of advanced computer technologies in drug discovery and development has been slow. The techniques that have been prevalent in engineering for many decades – modeling, simulation, analytics, machine learning and algorithmic predictions – had taken a second seat in drug R&D as educated guesswork and blind luck seem to have dominated (1). The reasons are unclear but it appears that for over two decades, scientists had too much confidence in their own abilities to navigate through the complex non-linear interactions in biological entities and pick winners and losers.

The results have been disastrous. Decision quality – the ability to consider all available information at the point of decision – has been dismal in this industry as human brains, albeit being logically superior to computers, are prone to biases and suffer from a lack of capacity to hold a large number of interacting variables. This is a tough message for those who have taken immense pride in moving R&D programs along by sheer intuition and experience. They made selection decisions based on a few observed characteristics and historical biases. They made design decisions based on a few industry standard templates and they largely shunned portfolio management as not useful. In the process, the economics of the R&D machine deteriorated under heavy attrition and cost overruns.

It is time to think differently and borrow heavily from other industries. The fact that pharmaceutical R&D is regulated and that it is dealing with systems not driven by engineering design principles are not sufficient reasons not to consider available tools. Continuing down the well-trodden path is not going to make things any better soon.

(1) Molecular matchmaking for drug discovery. Phys.Org. June 5, 2012