Skip to main content
Premium Trial:

Request an Annual Quote

GNS Healthcare, MMRF Bring Patient Simulations to Multiple Myeloma Precision Treatment, Trial Design


NEW YORK – What is the best strategy for testing new drugs in patients with relatively rare cancer types? Is there an optimal treatment combination or sequence that might benefit one patient but not another? Which tumor targets are most apt to respond to treatment?

GNS Healthcare is banking on artificial intelligence to answer these and other questions for patients with multiple myeloma, a rare cancer marked by aberrant clonal expansions of antibody-producing white blood cells, known as plasma cells, that develop from B cells in the bone marrow.

The Cambridge, Massachusetts-based company, focused on using causal machine learning to advance precision medicine, recently announced that it has come up with an in silico multiple myeloma patient tool called Gemini to model disease progression, treatment response, and other potential clinical outcomes — an approach that it brings to a multi-year collaboration with the Multiple Myeloma Research Foundation (MMRF).

"What Gemini is enabling is the ability to run as many clinical trials [as necessary], patient by patient by patient, to discover what works for who," explained Colin Hill, chairman, CEO, and cofounder of GNS Healthcare. 

As part of a five-year collaboration announced earlier this month, the firm is teaming up with MMRF to dig into whole-genome sequencing, exome sequencing, clinical outcomes, and other data generated on samples from more than 1,100 multiple myeloma patients enrolled in MMRF's Relating Clinical Outcomes in Multiple Myeloma to Personal Assessment of Genetic Profile study, or CoMMpass, while simulating multiple myeloma patient features to try to boost patient management, treatment decisions, and trial design.

For the longitudinal CoMMpass study, bone marrow biopsy samples are collected for sequencing from participants at diagnosis, as well as at points of disease progression. MMRF banks the samples with the Mayo Clinic in Scottsdale, Arizona, and makes the data from the study available to other investigators through an open-access database known as Researcher Gateway. 

The study has already started to clarify the genomic landscape of multiple myeloma by highlighting disease subtypes that influence outcomes and treatment responses, explained Michael Andreini, MMRF's chief operating officer.

Last summer, MMRF launched an Immune Atlas precision medicine pilot program to complement that data. For that project, investigators at Beth Israel Deaconess Medical Center, Emory University, the Mayo Clinic Rochester, Mount Sinai School of Medicine, and Washington University are using assays such as single-cell RNA sequencing and mass cytometry single-cell profiling (CyTOF) to profile immune cell populations and immune responses in samples from CoMMpass participants.

But investigators at MMRF and GNS Healthcare believe there is still much more to learn from these and other growing multiple myeloma datasets.

"We've invested in this great asset," Andreini said. "How can we bring to bear some of these advances — like AI, machine learning, deep learning — to help to answer some of these questions that our patients have, and, in particular, start to generate hypotheses and potentially validate hypotheses that will guide how patients should ultimately be treated, in a precise way, depending on their particular genomic, immune, and clinical make up?"

Available or proposed multiple myeloma treatments have mushroomed to include drugs targeting CD38 or SLAMF7, proteasome inhibitors, immune-based therapies such as CAR-T cell therapy, and others, he explained. But despite growing treatment options, there are still multiple myeloma patients who do not respond to existing drugs and are in need of new strategies.

"With the data that we've been generating … and with the application of some of these advanced modeling techniques, you can start to inform how to more precisely deliver the therapies that are available and also help inform the discovery of additional targeted therapies that can be explored in the future," Andreini said.

The MMRF has also undertaken a trial called MyDRUG that matches multiple myeloma patients to targeted treatments approved in other cancer types based on testing for half a dozen specific alterations, explained Anne Quinn Young, chief marketing and development officer at MMRF.

"There's a potential development path for each of these drugs for myeloma — or, at least, the potential to gain enough evidence that they could be prescribed and reimbursed off-label," she said, noting that a small subset of multiple myeloma patients with BRAF V600E mutations discovered in an earlier study have shown promising responses to off-label treatment targeting that alteration.

The hope is that simulated clinical trials will build on data produced by such earlier research and inform future precision oncology efforts in multiple myeloma, which has fewer targeted treatment options than some solid tumors, such as breast or lung cancer.

While simulations showing how a system responds to a range of conditions are common in other industries, Hill said, it has taken time to apply them to the clinic or to human biology more broadly, in part because "we didn't even have the full parts list," or blueprint, of how normal or diseased human samples are put together and the activity of the interacting parts behind these systems.

That started to change with the advent of human genome sequencing, and as gene expression, proteomic, and other molecular tests began being increasingly used to study the features of healthy and malignant biological samples from individuals with multiple myeloma and other cancers.

Such technologies "gave rise to the ability to now measure, across a number of patients, multi-modal molecular activity that can be matched up with both drug treatments and clinical outcomes," Hill said. "Now, you have the building blocks to potentially make sense of, and reverse engineer and unravel, how these systems work."

Building on data generated over several years, the Gemini tool brings together clinical and genomic profiles for patients with supercomputing, and applies analytics within a causal AI framework, to "try to actually unravel what's causing what," he noted, rather than looking for more general correlations.

Because CoMMpass is longitudinal, Hill noted, the team expects to see some cases that progress or become treatment-refractory over time, as well as patients who are receiving targeted therapies, or a sequence of treatments, that were not on the market when the effort began.

The patients enrolled in CoMMpass also appear to be quite representative of the broader multiple myeloma patient population in the US, where men — particularly men with African American ancestry — and elderly individuals bear much of the brunt of the disease incidence.

And for an AI application such as Gemini, capturing this patient and treatment diversity will be crucial, Hill said, since the tool needs to make as many relevant linkages as possible to accurately predict characteristics and treatment responses for newly diagnosed multiple myeloma cases.

With that in mind, the Gemini investigators will continue to incorporate additional publicly available datasets and data from other collaborators, as well as look at commissioning the generation of new data.

"If you're trying to figure out how the variation across the genome or gene expression patterns is impacting outcomes, or [how] survival changes, under different drug treatment conditions, you need that kind of variability," he explained, noting that if the patients, their tumor features, and treatments were homogeneous, "we wouldn't be able to learn a thing."

To avoid concerns about clinical decision support recommendations stemming from a closed, black-box system, Hill said, the team has made Gemini as transparent as possible, so that users can retrace the rationale behind specific management decisions recommended by the simulations.

"In this space, it is super important that the mechanistic reasoning is there because of course, these are life-and-death decisions and treatments," he said, "so one needs the traceability to say, 'Why is it that drug X works better than drug Y?'"

Along with efforts to come up with accurate tools to guide treatment decisions in the clinic, GNS Healthcare is exploring Gemini as a tool that drug developers can use to inform their R&D pipelines and treatment commercialization strategies. For instance, in silico patient-centered strategies can unearth drug candidates with potential activity in tumors with specific features, and can conduct head-to-head simulations with other drugs to generate data to support applications for regulatory approval.

Andreini also touted the potential benefits of using in silico patient models to design clinical trials for specific patient populations or treatment strategies, depending on a drug target or mechanism of action.

"If [companies] can be more strategic in how they approach the evaluation of various combination approaches or [treatment] sequencing approaches … and if they can develop hypotheses to better enrich patient populations for trials," he suggested, "that allows them to potentially accelerate the speed at which they conduct those trials, because you have a better chance of seeing a signal that's statistically meaningful and clinically meaningful."