Peter Jarowski

Chemical Data Has Problems

The state of data access, quality and dissemination in Chemistry is extremely poor - so poor that it is blocking advances in machine learning (ML) and artificial intelligence (AI), and also impeding research and development in traditional methods. The recent surge in AI skepticism is a direct consequence of years of over-hype and promises based on precarious data. Over-the-top expectation were offered without enough consideration for the data quality and volume required to train fancy algorithms. The old adage “^&$% in, ^&$% out” holds true (we can say ‘crap’ right?). This opinion is in line with recent statements by the CEO of Novartis, for example, who runs the second largest pharmaceutical company in the world, lamenting the difficulty in accessing quality datasets to make AI effective.

Since August Kekulé’s proposal for the tetrahedral configuration of carbon or his more famous realization that benzene was a cyclic molecule, a snake biting its tale, molecular structure has been the leading consideration for the design of new molecules as drugs or performance materials. For the former, it is said that 70% of drug design is based on molecular shape with the remainder attributed to electrostatic or non-bonded interactions.

Structural chemistry began around the 1860 with these dual assignments by Kekulé but it wasn’t until one hundred years later with Allinger’s initial force field approaches that the first classical molecular mechanics (MM) models became available to make computer-assisted prediction of molecular structure. These models themselves are based on principles derived by Robert Hooke, a contemporary of Isaac Newton, in the mid 17th century with additional layers from van der Waals (19th century) etc.

In the domain of drug discovery, there can be a world of difference between a computer-generated hit compound, which is predicted to bind well to a drug target and what can be reliably synthesized at scale, or indeed synthesized at all. This discrepancy has been a lingering point of discord between the Discovery and R&D efforts in the chemical industry. Computer-aided drug design (CADD) has become an increasingly valuable tool by providing essential screening data and unique insight into drug action and mechanism, but it does not model the more complex world of chemical reactivity and synthetic chemistry.