Machine Learning and Drug R&D

machinelearningWe recently wrote about the growing use of machine learning and artificial intelligence (AI) in healthcare. That blog post talked primarily about the expanding applications of AI in clinical diagnostics, based on AI’s superior abilities in pattern recognition. Within the pharmaceutical industry, however, machine learning technologies hold the most potential for revolutionizing drug discovery and development.

 

Machine learning is technology based on the building of a mathematical model that can use large data inputs to predict something, solving problems or conducting experiments that a person cannot normally accomplish.

 

For example, researchers at the University of Cambridge recently published a machine learning algorithm in Proceedings of the National Academy of Sciences (PNAS) that enables the identification of potential new drug compounds with twice the efficiency of traditional drug discovery methods. The researchers used the algorithm to analyze molecules known to be active or inactive against a certain target, teaching the algorithm to recognize which parts of the molecules are important or unimportant for drug activity, and to identify important chemical patterns. This enabled the researchers to gain insights about drug structures and functions from both successful and failed experiments, using that knowledge to predict possible novel compounds with activity against the target of interest.

 

Starting with 222 molecules known to be active, the researchers built a model and used it to computationally screen 6 million more molecules. They then tested the 100 most relevant molecules picked out by the algorithm, ultimately identifying four previously unknown molecules that activated the drug target of interest.

 

Pharmaceutical developers are already teaming up with academics and AI developers to employ machine learning approaches in their own discovery and development efforts.

 

For example, the departments of Chemical Engineering, Chemistry, and Computer Science at the Massachusetts Institute of Technology (MIT) have formed the Machine Learning for Pharmaceutical Discovery and Synthesis Consortium with a group of seven pharmaceutical companies. The Consortium’s goal is to facilitate the design of software to automate small molecule discovery and synthesis.

 

Charles River and AI-specialist Atomwise have signed a $2.4 billion collaboration focused on streamlining early-stage drug discovery using Atomwise’s technology to predict how well a small molecule will bind to a protein target of interest. This approach enables scientists to computationally test a large and diverse number of chemicals within just a few days, streamlining the drug optimization process.

 

GlaxoSmithKline and 23andMe have similarly formed a multi-year collaboration to identify new drug targets, disease subsets, and better methods of stratifying patients for clinical trials based on the use of machine learning methods to predict individual disease risk from genetic, health and lifestyle data.