Skip to main content
Premium Trial:

Request an Annual Quote

GNS Healthcare Expects Increased Biopharma Interest in Causal Machine-Learning Tech


NEW YORK (GenomeWeb) – GNS Healthcare and Celgene have signed a service and license arrangement that allows the biopharma firm to use GNS Healthcare's causal machine-learning and simulation platform for applications in drug discovery, clinical development, and commercialization and market access.

Celgene has also made a second equity investment in GNS. Last year, Celgene was one of several investors that participated in a $10 million Series C financing round during which GNS raised $10 million that it used to, among other things, improve its machine learning-based platform. Under the terms of the licensing agreement, an undisclosed number of several causal modeling experts from GNS will be brought in-house at several Celgene sites to operate the platform. 

This arrangement has not been the typical way the company has worked with collaborators in the past, but requests for more access to GNS's technology have been growing as the appreciation for applying breakthroughs in machine learning becomes more apparent, according to GNS Chairman and CEO Colin Hill. While most biopharmaceutical companies, Celgene included, have trained bioinformaticians and machine learning experts in house, they may not be trained in the particular approach that GNS uses.

"There are different kinds of machine-learning and artificial intelligence technologies and approaches and internal software," Hill told GenomeWeb. "People are [now] starting to understand how to match the right type of technology within machine learning to the right types of problems."

Given the different kinds of machine-learning and artificial intelligence technologies available in the market, it makes sense for companies like Celgene to bring outside experts in to help them apply the right technologies to solve their problems. "The causal machine learning approach that GNS uses is one specific kind of machine learning, [so] it really makes sense to operate in this way," Hill said.

Cambridge, Massachusetts-based GNS Healthcare offers services and products based on its so-called reverse engineering and forward simulation (REFS) platform, which uses patented machine learning and simulation algorithms to identify causal relationships in diverse datasets and build predictive models based on these relationships that help users answer specific questions. It uses a combination of genomic, molecular, clinical, pharmacy and medical claims, electronic medical records, and other datasets to generate models that help users explore "what if" questions, such as how a patient might respond if they receive a particular treatment. Customers use the platform to predict individual and population risk and costs, compare the effectiveness of treatments, improve medication adherence, and identify biomarkers that predict how patients respond to treatment.

Specifically, GNS' technology helps users establish causality, according to Hill. A lot of machine learning focuses on finding patterns and correlations in data, so, for example, "what happens when I see a level of this blood biomarker what do I see [elsewhere]?" he explained to GenomeWeb. "That's not the same as being able to determine what happens when I intervene [or] what happens when this patient takes this diabetes drug at this dose? That is what our technology enables."

For example, the Multiple Myeloma Research Foundation tapped GNS's platform to explore genomic and clinical data from the Relating Clinical Outcomes in Multiple Myeloma to Personal Assessment of Genetic Profile, or CoMMpass, study. The partners shared some discoveries from that result this week. Among other findings, they identified CDK1, PKMY1, MELK, and NEK2 genes as the top drivers of high-risk disease. These genes represent a pathway that contains known drug targets, suggesting a validation strategy and the potential to employ drugs in combination, the partners said. 

GenomeWeb reached out to Celgene for additional details on how the company will use GNS's technology by was unable to get a response as of press time. However, according to GNS, the platform could support efforts to develop new drug targets and identify pathways that if targeted may reduce tumor size and extend progression-free survival of cancer patients. The second important application for pharma is identifying which subset of patients will respond positively to newly developed drugs and which patients will have an adverse event.

Understanding causality is crucial to both of those efforts, Hill said. He pointed to recent reports that an experimental Alzheimer's drug called solanezumab, developed by Eli Lilly, failed in a phase III clinical trial involving more than 2,000 patients — the drug targets the amyloid-B peptides that form plaques in the brain. "There are number of reasons why the drug may not have worked, but one likely reason is it's not the right drug target," Hill said. The pathway by which the plaques are formed are likely correlated with the disease but not a causal driver. 

Studies like these are potential use cases for machine-learning technologies like those offered by GNS, Hill said. Last year, the company announced a partnership with The Alliance for Clinical Trials in Oncology that sought to identify patient subpopulations that are likely to respond to combination treatments for metastatic colorectal cancer. Under the terms of that agreement, GNS will use its platform to analyze genetic, genomic, and clinical datasets from a recently completed Phase III clinical trial of more than 1,100 patients to evaluate the efficacy of a series of combination chemotherapies.

Hill does not expect that GNS will embed its employees at every location like it is doing with Celgene, although that is one route the company will take to give customers greater access. A lot of those decisions will depend on the customer in question and what their needs are as well as what sort of expertise they have in house. " We really created it to be flexible ... and to essentially meet them where they are," Hill said.

Meanwhile, GNS continues to partner with customers to develop custom models for specific needs. Existing partnerships include one signed in 2013 with the Inova Translational Medicine Institute to develop models for predicting preterm live birth risk based on genetic and molecular factors with clinical, environmental, and behavioral data as well as health outcomes. Last year, the company was listed as one of the founders of the non-profit Transforming Medicine: The Elizabeth Kauffman Institute (TMed), which seeks to match patients with life-threatening diseases with the most appropriate treatments. TMed plans to develop a knowledgebase for evaluating patient information, including genetic and molecular data, as well as develop algorithms that can predict the most effective therapies for difficult-to-treat ailments including pancreatic cancer, brain cancer, and sickle cell anemia.

In preparation for its expected growth, GNS is adding staff and further developing its platform using funds from its most recent financing round, Hill said. These efforts include automating processes within the REFS platform as well as putting infrastructure in place to run computer models — which incorporate genomic sequence, mobile health data, gene expression data, proteomics data, and more — more efficiently. "We were early cloud adopters and early users of the non-cloud supercomputers, so luckily the hardware part has really been less and less of our direct problem," Hill said.

The company is also creating derivative products using its technology and data that the company has in-licensed or partnered to obtain. "We actually run data from large populations through our technology platform to ... enable us to solve certain problems," Hill said. For example, one application that the company has recently begun offering targets customers, such as health insurance companies, working in the palliative care space.

Last year, the company launched Max for Metabolic Syndrome, a population health management solution that predicts which of five risk factors associated with metabolic syndrome will develop in at-risk individuals. It also matches each individual with the most effective treatment aimed at preventing the condition from developing.