Skip to main content
Premium Trial:

Request an Annual Quote

CytoReason's Machine-Learning Model Offers New Insights Into Immune-Related Diseases


NEW YORK (GenomeWeb) – Informatics firm CytoReason is betting that its method of modeling immunological activity could be a boon to pharmaceutical companies that are looking to identify potential drug targets or diagnostic biomarkers from multi-omic and immune interaction data gleaned from private and proprietary sources.

Founded in 2016, CytoReason's platform utilizes supervised and unsupervised approaches to model molecular, cellular, tissue, disease, and drug relationships in a manner that supports target and pathway identification, and improved indication selection and prioritization. The platform can also provide insights into adverse events as well as identify preliminary safety and off-target effects.

The platform can help pharma companies stratify patients based on response versus non-response, according to CytoReason. A recent paper, published in the journal Gut by CytoReason in collaboration with researchers at Stanford University, Tel Aviv University, and elsewhere demonstrated this capability in the context of inflammatory bowel disease cases. In that paper, the collaborators use the algorithms to identify a new blood-based biomarker that is 94 percent accurate in identifying non-responders to anti-tumor necrosis factor alpha (anti-TNFα) therapies in IBD cases.

Non-responders to anti-TNFα therapies such as infliximab account for approximately 30 percent of IBD cases, but to date there have been no predictive assays for identifying non-responders, the paper noted. Specifically, the collaborators found a correlation between the proportion of plasma cells present and non-response to therapy in two independent cohorts of cells gleaned from whole-genome expression profiles of colon biopsies. Their analysis showed a rise in inflammatory macrophages associated with increased expression of two receptors in the TREM1-CCR2-CCL7 axis, which is responsible for inflammatory monocyte movement, in non-responders. Upregulation of this pathway in patients is predictive of non-response to treatment with infliximab.

The researchers then validated their results in a fresh set of biopsies and blood samples gleaned from three separate test cohorts.

In addition to highlighting the need for more effective therapies for IBD patients, these results suggest at least two possible clinical assays for predicting non-response to anti-TNFα therapies prior to starting treatment, according to the paper. For example, a potential assay might use expression of the TREM-1 receptor as a non-invasive diagnostic of non-response to anti-TNFα therapy while the two receptors identified in the studies could serve as therapeutic targets for patients with refractory IBD that could guide the development of more effective therapies.

More importantly for the firm, these results provide proof of CytoReason's ability to find previously unknown patterns in omic measurements or yield new insights into understudied patterns, according to the company.

"We demonstrated our ability to recreate and separate out cellular contributions from bulk tissue measurements, discovering a pretreatment upregulated pathway uniquely in anti-TNFa non-responders," Shai Shen-Orr, CytoReason's chief scientist and director of the systems immunology & precision medicine laboratory at The Technion - Israel Institute of Technology, said in a statement. "Furthermore, we established a clear biomarker relationship between biopsy and blood, which we were able to validate with high accuracy."

These results also suggest that current methods of analyzing gene expression measurements may not provide the most useful signal, Shen-Orr explained in an interview. "When people look at gene expression, they actually are looking at information that is very clouded." That's because variations in the proportions of cells present may mask important signals. For example, in the IBD study in Gut, non-responders had more inflammatory cells in their biopsies and as such showed greater levels of expression.

"Once you realize this, and you take this into account, you can unmask the signal where the people who are non-responsive actually have an upregulated pathway that you wouldn't have picked up otherwise," he said. "The biology becomes much more easily interpretable as soon as you put it into this cell-centric model that CytoReason is developing."

CytoReason's technology uses a proprietary data and machine-learning model to reconstruct cellular information from bulk tissue, to train an immune-specific natural language processing engine, and to integrate multi-omics data. The platform accepts gene expression, cell and protein measurements, flow cytometry data, microbiome data, as well as peer-reviewed results from the public domain and immune interaction data, among other sources as input. It integrates it into a proprietary disease model to generate a mechanistic understanding of the immune system.

The underlying technology on which the platform is based was initially developed at Stanford University during Shen-Orr's post-doctoral studies at that institution. "The basic idea is to try and reason over heterogenous cell populations," he explained. He pointed to the millions of omic datasets from microarrays and next-generation sequencing instruments, noting that though the data-generation technologies have improved drastically, the ability to glean usable information from the data is still in its nascent stages.

In addition to the work described in Gut, Shen-Orr and his collaborators have published papers that demonstrate other components of the CytoReason platform. For example, members of his lab at Technion published a separate paper in Nature Methods, which described a framework for generating trajectories of biological processes such as cell differentiation or cell development from single-cell gene expression data or flow cytometry data. It is the first algorithm of its kind that provides access to that level of information from single-cell data, according to Shen-Orr.

"We are in this revolution where the measurements are easy but actually figuring out what all this data means and giving the bang for the buck is not trivial," he said. CytoReason was set up to commercialize "a bunch of different technologies that we developed to get the utmost from immune data," he said. "This is really an unmet need both in academia and in industry."

The CytoReason platform shares some similarities with technology developed by researchers at University of California, San Diego and Tel Aviv University. Specifically, those researchers have developed an in silico model of a yeast cell, called DCell, that simulates cell growth from genotype information. Where the two systems differ, Shen-Orr explained, is that while DCell focuses on modeling cellular behavior, CytoReason's platform focuses on modeling the immunological activity.

CytoReason offers services based on its platform to pharmaceutical companies and research institutes interested in identifying new treatment targets, biomarker selection, or understanding the mechanism of action of particular drugs. Shen-Orr said that company has already worked with several large pharma companies who have used its services along those lines. In addition, the company is involved in a partnership with researchers at the Parker Institute for Cancer Immunotherapy, which aims to develop improved treatments for cancer.

"Our mission is to accelerate some of the breakthroughs we are seeing in immunology and get them out faster, and a key component of that is understanding the dynamics of the immune system," Nikesh Kotecha, vice president of informatics at the Parker Institute, said in an interview. "In order to really do that, you have to actually be able to measure data from different modalities. [Once] you have that information, how are you going to [put] it together, and what does that mean in terms of how you put it together and what is the context around that?"

The technology that CytoReason has developed is designed to help researchers ask those types of questions, he continued. For example, in terms of analyzing data from responders versus non-responders, "what specifically about these immune features is driving those pieces and what technologies are helping you measure those pieces … and then how do you use [that information] to optimize what you measure?" Some of the work that CytoReason is doing "helps you get a sense of that," he said.

Based in Tel Aviv, Israel, CytoReason currently has about 20 employees on staff and is looking to expand its headcount in the coming months, Shen-Orr said. The company also plans to take on more clients over the next several months. Its business model is to form long-term engagements with potential clients where there is a possibility for mutual benefit.

Under this model, the company does not own any of the intellectual property that results from its client's research. Instead it receives payment upon completion of pre-established milestones and other deliverables, he explained. The particulars of each engagement vary from one project to the next. Furthermore, longer-term partnerships make more sense for the firm, because the more data the system is exposed to the better its predictions are over time, and as project partners generate new datasets, the CytoReason platform can generate better, more precise models.

CytoReason is also looking to make improvements to its platform that will allow the company to incorporate more data types, such as clinical trials data, as well as support new applications of its models within the immunological domain. For example, while the company is currently focused on autoimmune diseases and immune oncology, it hopes to expand into neurodegenerative diseases and infectious diseases.

"We think the technology can actually be leveraged across many diseases." However, "we are prioritizing [diseases] where there is a clear understanding that the immune system actually plays a role," he said.