Machine learning answers 'holy grail' questions to accelerate drug development
The service and license arrangement with GNS Healthcare (GNS) gives Celgene Corporation rights to operate the GNS Healthcare REFS™ (Reverse Engineering and Forward Simulation) causal machine learning and simulation platform.
As part of the arrangement, several GNS causal modeling experts will be brought in-house at Celgene sites to operate the platform. Additionally, Celgene has made a second equity investment in GNS.
The core GNS Healthcare technology, REFS™, is “fundamentally different from all other types of machine learning approaches,” said Colin Hill, CEO, chairman and co-founder of GNS Healthcare.
“Causal modeling and simulation is the only type of technology capable of answering the ‘holy grail’ questions that are necessary to better match drugs and other health interventions to individual patients and discover new pathways for intervention,” he told us.
According to the company, the platform leverages combinations of genomic, molecular, clinical, pharmacy, and medical claims, EMR, emerging real-world, and other types of data to reveal causal mechanisms between variables.
“Causal modeling and simulation can discover causal targets and pathways from the near infinite possibilities—like finding a needle in a haystack—which cannot be done otherwise,” added Hill.
As Hill explained, the technology simulates experiments computationally, rather than in the wet lab, which exponentially increases the number of experiments that can be performed. It can also answer complex questions, such as what treatment will work, at what dose, for a particular patient.
The technology will be applied across the pipeline, from drug discovery to clinical development and health economics and outcomes research (HEOR).
Machine learning in pharma
Hill said that the company’s platform often gets compared to that of IBM Watson or Google – but he stressed that the technology is different.
“When REFS consumes data, it uses induction to infer the mechanisms that gave rise to the data. If you can solve that problem, you can use simulation to answer all sorts of questions about the impact of interventions, patient by patient,” he explained.
“Some of the most sophisticated machine learning operations in the world, like Amazon, rely on predictive models to power their recommendations.”
However, predictive models are not causal models, as predictive models look at data to find patterns, correlations, and calculate probabilities.
Hill also provides the example of IBM’s Watson, which uses deduction and is based on driving conclusions from existing human knowledge.
“These approaches cannot determine what is causing an outcome and, therefore, cannot determine which treatment will have the greatest clinical impact for a given patient,” he said. “They have an important place in the toolbox, and they are also helpful to illustrate how causal modeling and simulation is so unique and powerful.”
As Outsourcing-Pharma.com. previously reported, IBM Watson cognitive computing is helping match patients with clinical trials. Companies, such as Icon Clinical Research, have adopted the technology to accelerate the traditionally inefficient process.
Last week, Pfizer also announced a collaboration with IBM Watson Health, in which the company with use IBM Watson for Drug Discovery to help accelerate research in immuno-oncology.
Pfizer has also worked with IBM to develop remote monitoring solutions.
