AI WILL EMPOWER DRUG DISCOVERY, BUT NOT THROUGH IN SILICA EXPERIMENTATION
Artificial intelligence, or “AI” for short, has been the topic of sweeping news coverage of late. In the life sciences arena, much focus has been given to the use of AI to help the drug discovery process. Articles like “Big pharma turns to AI to speed drug discovery, GSK signs deal” and “Artificial Intelligence In Drug Discovery: A Bubble Or A Revolutionary Transformation?” are both examples of media coverage speculating on whether the use of AI can significantly impact drug discovery efficiency. From where we sit, we tend to think the answer to such questions is generally yes, but also tend to think it will not be in the way that many people think.
To be sure, AI will have a major impact on life science research. Properly configured, AI has a very strong chance of analyzing large data sets for genetic and population characteristics to identify the root causes of disease, and potential targets therein. On the other hand, much of the capital being poured into AI development for drug discovery is directed toward the use of AI in predicting the way molecules will interact before experimentation, in effect attempting to move experiments out of lab ware and into a virtual experimentation mode. This so called “in silica” testing via AI-empowered computational chemistry is fascinating, but in our opinion, a long, long way from being able to replace true experimentation. The reactions are complicated, and while advances in computational chemistry and AI more generally are encouraging, we tend to think the complexity and dynamism will make accurate predictions a challenge.
On the other hand, productivity improvements from laboratory automation systems is both driving down the cost of experimental data, and producing a ton of information – more information in fact than most scientists can handle. If you look at a modern robotically driven high throughput screening system with cell imaging capability, the incremental unit cost per image has dropped quite a bit, but the number of images being produced is now quite large. Rather than using AI to predict experimental results in advance (which we believe will prove very hard), we believe there is a great opportunity to use the power of AI to more efficiently analyze experimental data already being created. Imagine a highly efficient AI scanning cell images like a digital pathologist, and alerting the scientist for shared features and characteristics tied back to experimental attributes. Powerful!
If AI could focus on downstream data analysis, automation can focus on upstream productivity gains. Together, they could be a powerful force in reducing discovery costs.