Machine Learning Could Accelerate Drug Discovery 

According to the Congressional Budget Office, in 2019, the pharmaceutical industry spent $83 billion on research and development, about ten times the amount spent per year in the 80s. With so much money going toward exploring whether or not a drug will be effective, one group has turned to machine learning (ML) to help expedite the process.

Researchers at Massachusetts Institute of Technology (MIT) shared the results of their machine learning technique, called DeepBAR, and how it is positioned to help better determine the effectiveness of new medications. The study's lead author is Xinqiang Ding and the co-author is Bin Zhang.

“Drugs can only work if they stick to their target proteins in the body,” writes Daniel Ackerman, for MIT and as published by Science Daily. “Assessing that stickiness is a key hurdle in the drug discovery and screening process.” 

Enter DeepBAR, which can quickly determine how likely drug molecules are to bond to target proteins – known as the binding affinity—which is difficult to determine using traditional methods.  

“The affinity between a drug molecule and a target protein is measured by a quantity called the binding free energy,” says Ackerman. “The smaller the number, the stickier the bind.” 

This means that drugs with a lower binding free energy will have more success competing against other molecules, interrupting the protein’s normal function. Knowing this information is one of the biggest indicators of a medication’s effectiveness, says Ackerman. But determining that number is a cumbersome process using existing methods. 

Of the current two approaches, one requires a large amount of computational power and time to complete while the other uses less of both but is also not as accurate. 

DeepBAR takes traditional chemistry methods and combines them with machine learning—specifically the “Bennett acceptance ratio, a decades-old algorithm used in exact calculations of binding free energy,” says Ackerman. 

The result is calculations of binding free energy on small protein-like molecules were completed 50 times faster than prior methods. Work continues on using DeepBAR on large proteins. The hope is that in the not-so-distance future, DeepBAR will be ready for use in screening of new medications and could even potentially be used for new medications used for the treatment of COVID-19.

“In the future, the researchers plan to improve DeepBAR's ability to run calculations for large proteins, a task made feasible by recent advances in computer science,” concludes Ackerman.

The study is yet another example of how the computer sciences, particularly advances in machine learning, artificial intelligence, and natural language processing, have wide-reaching impacts across a multitude of industries, including healthcare.  

For more information on DeepBAR, refer to the full study, published in the March issue of the Journal of Physical Chemistry Letters.

Capitol Tech offers bachelor’s, master’s, and PhD programs in cyber analytics and data science, including courses in machine learning, artificial intelligence, and natural language processing. Email admissions@captech.edu for more information.