Deep learning’s deepening impact on drug discovery and free energy simulations in industrial computer-aided drug design. What our chemists are looking for in the coming year.

Artificial Intelligence in Drug Discovery

The idea of AI has been around for decades but the combination of big data, powerful computers and smart algorithms is bringing about a resurgence of interest in drug discovery applications.

I wrote earlier this year about the emerging use of Deep Learning in drug discovery. The rise of artificial intelligence techniques applied to pharmaceutical research continues apace and the trend can exemplified by two very recent news items:

  • On Nov. 10, IBM Watson signed a five-year collaboration with the Broad Institute of MIT and Harvard University during which IBM researchers will apply Watson to examine genome data from 10,000 drug-resistant tumors and cell line studies with the goals of understanding the drivers behind cancer drug resistance and then developing new treatments.
  • On Nov. 8, the UK company BenevolentAI announced an agreement with the pharmaceutical giant Janssen under which BenevolentAI obtained access to a select number of Janssen’s novel clinical stage drug candidates and extensive related portfolio of patents. This selection was based on work by BenevolentAI scientists,who applied the company’s AI technology to evaluate the potential of the Janssen molecules to become new medicines for hard-to-treat diseases. 

These represent just a brief snapshot of a burgeoning field which looks like a hot topic for 2017. For further reading, check out these announcements by IBM, and BenevolentAI, and these articles by Inverse, Pharmafile, CB Insights, and Chem Commerce.

—David Clark, Research Fellow in Computer-Aided Drug Design (CADD

and Information Services and founder of Argenta


Free Energy Perturbation

Once a technique predominantly utilized and developed by academics, the widespread adoption of free energy simulations into industrial computer-aided drug design is now on the horizon.

The holy grail of in silico techniques is to be able to predict how tightly an inhibitor will bind to its target. Computational chemists, largely based in academia, have grappled with this for decades, devising numerous so-called ‘free-energy’ techniques in the hope of accurately predicting ligand binding. One such technique is Free Energy Perturbation (FEP), which has had something of a renaissance in the last couple of years.

FEP seeks to calculate the difference in binding affinity between two ligands rather than their absolute affinities; something which is considerably more computationally efficient and also more reliable. The adoption of FEP into commercial software packages such as Schrödinger’s FEP+ represented something of a watershed in the field, allowing CADD scientists to use these tools in earnest on their own targets.

Like any computational method, the results obtained are always limited by the description of the underlying physics. With the adoption of the method into industrial pharmaceutical research, however, there will inevitably be a drive to improve these descriptions, and as such it is likely that the use of FEP will only increase in the coming years.

—Michael Bodnarchuk, PhD, molecular modeler

Discovery Research Services