6:16 PM
 | 
Jan 20, 2017
 |  BioCentury  |  Tools & Techniques

Hopes in the machine

How machine learning is being used in drug R&D

Applying machine learning tools to drug R&D could usher in a new era of effective therapies targeting causative disease drivers, higher success rates in clinical trials and real-world tools to manage adherence, prevent adverse events and reduce total healthcare costs.

Biopharma companies have been working at mining big data since the genomics era began, but traditional data mining parses only one type of data at a time to yield correlations. For example, mining sequencing data can reveal a correlation between a patient population and a disease target or mutation. But it doesn’t reveal whether the mutation is a causative agent in pathogenesis, or its relationship to other factors that affect the safety and efficacy of targeted drugs in different patients.

In contrast, machine learning tools analyze multiple sets of data from disparate sources -- often in wildly different formats -- to sniff out connections among biology, genetics and phenotypic traits.

Using an iterative, learning approach, the technologies have the potential to identify previously unknown pathways, uncover which targets are causing rather than simply correlated with a disease, predict how patients will respond to therapies, and draw conclusions about whether a clinical trial will be successful or how a drug will perform in the real world.

These technologies are created from a series of neural networks -- a computational technique that can be used to discover and predict patterns among a set of inputs and corresponding outcomes.

They employ natural language processing to analyze unstructured data from the literature and physician notes in electronic medical records (EMRs), and combine this information with millions of structured data points to develop hypotheses that can then be tested in preclinical animal models, clinical trials or the real world.

“Our goal is to be at least fourfold more efficient.”

Jackie Hunter, BenevolentAI

Adoption of these technologies remains modest but could increase over the next few years as early programs deliver successful clinical trials, approved drugs and better outcomes in the real world.

A compound developed for a cancer population selected using GNS Healthcare Inc.’s reverse engineering forward simulation (REFs) technology could be on the market within a few years. The company also has shown its Efficacy to Effectiveness (E2E) tool can accurately predict real-world performance of a new drug prior to its launch.

Two companies using their platforms to discover targets and drug candidates, BenevolentAI and Berg LLC, plan to have compounds in the clinic in the next year.

Many, many more experiments are ongoing via partnerships as drug developers and providers seek to understand whether and how these tools can save time and money, and improve research and clinical outcomes.

“I think value-based arrangements with machine learning tracking outcomes will be the sharp ends that I think will pay off very quickly within the next year, for sure,” said John Glasspool, head of corporate strategy and customer operations at Shire plc, which is evaluating a handful of undisclosed machine learning technologies.

As these tools start to deliver improvements in patient adherence and identification of specific patients who will benefit from a drug, “we will get a better understanding of the disease biology and targets” he added. That in turn will result in greater Phase II and Phase III successes, he said.

Realizing that potential, however, will require much more data sharing than is currently taking place, along with creation of new data standards (see “Driving Adoption”).

Profiles of nine companies who spoke to BioCentury and are developing machine learning technologies for use in drug R&D, from discovery to real-world evidence development, follow.


Sidebar: Driving adoption

Read the full 4195 word article

User Sign in

Trial Subscription

Get a 4-week free trial subscription to BioCentury

Article Purchase

$150 USD
More Info >