Supporting Drug Discovery with Cell-centered Models of the Immune System

Many biological processes, and specifically the immune system, operate at a cellular level. CytoReason have harnessed this knowledge to build computational, high-resolution, mechanistic models for a range of diseases that enable better discovery and validation of drug targets.

We recently spoke to David Harel, CEO at CytoReason, to learn more.  

Laura Elizabeth Lansdowne (LL): Could you tell us about CytoReason’s work towards better understanding disease and supporting more effective drug discovery and development?

David Harel (DH): CytoReason builds computational, high-resolution, mechanistic models on a cell-protein-gene level for a range of diseases and associated tissues. We marry those to treatment models for those diseases which we create from data we collect and from collaborator data. This platform allows us to rebuild lost cellular information from gene expression data, decouple the disease signal from the infiltrating cells' signal and enables us to associate particular genes to specific cells. This information is then integrated with additional omics and literature data to create a cell-based model of the trial-specific immune response. Integration with the CytoReason disease model empowers the study analytics and allows the model to learn and improve, leading to robust target discovery, drug response biomarkers and indication selection. 

In short, the platform enables us to: Better understand trial results (i.e. tissue/disease-specific cell/gene signatures); understand biological relations and signals outside the trial results; and to compare the effects of a single trial to other diseases and tissues.

This is really useful for: 

• The discovery and validation of targets, where looking into the relevant cells and associated genes allows us to identify or in silico validate genes of interest.
• Identification of variance within each disease model – enabling segmentation of patient populations at baseline to subgroups that are significantly different – allowing for discovering new targets with higher probability of success.
• Understanding collaborators’ trials by looking inside the cells for clues that are common among response groups – discovering response markers. The tissue model can then be compared to the blood model (same disease, different tissue) and be used to identify a blood marker, which is much easier to detect in a clinical setting.
• Comparing responder vs non-responder clinical data at cellular and gene level can yield predictive biomarkers which can be further utilized for patient stratification in advanced clinical studies 
• Comparing the effect of a treatment across multiple tissues and diseases, estimating off-target effects, and evaluating efficacy in other diseases for indication prioritization or expansion.

Ruairi MacKenzie (RM): Why does CytoReason use a cell-centered model rather than a gene-based model?

DH: Many tissues, and specifically the immune system, operate on a cellular level.  The cell is the base unit of many, if not most, biological activities. The benefit of using a cell-level, versus a gene-level model can be clearly seen in work that was published in GUT in 2018, where we analyzed inflammatory bowel disease (IBD) patients’ gene-expression data using our platform. When we evaluate the genes with the highest expression in our data, we see they mostly code for proteins and transcription factors involved in the infiltration of immune cells. That is no surprise, since there is inflammation and the infiltrating immune cells are infiltrating the infected tissue. The problem is that the abundance of these genes mask most other (potentially critical) disease signals, and these changes in immune response will result in a difference in gene expression proportion. The CytoReason Cell-Centered Models decouple the infiltrating cells'     signal from the disease signal, enabling a clear picture of important, yet hidden, pathways at play. This unique technology is overcoming the confounding immune response and resulting variance, which is of limited interest in this disease specific tissue. This work, which uncovered novel cellular and blood-based pre-treatment biomarkers, generated significant interest in the IBD field.

RM: CytoReason’s models leverage “inaccessible” data – what does this mean and how can it benefit drug discovery?

DH: We are living in a world that is rushing to integrate artificial intelligence (AI) and big data in every industry. The reality is that in healthcare and in pharma specifically, access to data is the main barrier, not the algorithms. Issues of privacy, intellectual property (IP), regulation and legacy systems make the utilization of data across pharma companies still rare – but gaining traction.

Every new collaboration and every new data set integrated into the CytoReason model enhances the strength and accuracy of the model – increasing the value to all collaborators. But it isn't simply raw data that makes the difference – patterns and context are the real keys to the accuracy of any machine learning platform. Our model requires only statistical summaries (correlations which cannot be reversed back to data) in order to grow – it doesn’t need the raw data, nor the questions or the answers asked of, and provided by, the model – this all remains totally confidential. Furthermore, once these summaries are integrated into the model, they also become indistinguishable from the rest of the model and cannot be separated out.   The integration process enabled CytoReason to establish itself as the trusted third-party among large pharma companies – a neutral party that can bring significant value to each collaborator while protecting patients’ information and collaborator IP. The key was to establish trust with the leading organizations and to maintain the highest ethical standards.

LL: What challenges and opportunities are associated with applying AI and machine learning to the drug discovery field?

DH: Clearly, there are always a myriad of challenges to overcome when a nascent field starts making serious inroads. We are occupied by three key challenges (in no particular order):

1. The "Black Box" issue: Drug discovery and development is a biological process – throwing data into a bunch of algorithms and then talking deep bioinformatics to a group of clinical researchers will not end well. This is true, both in terms of their understanding of the results and the trust they place in those results. Our team of biologists and bioinformaticians work hand in hand with the internal pharma research in order to bridge the gap between data science and biology. Each step in the process is explained and discussed with our collaborators, solving the “black box” uncertainties. The process focuses on delivering biological insights which the pharma research team instinctively understands and can further investigate.

2. Access to data: Commercial sensitivity, regulatory and privacy concerns, together with global cybersecurity threats are a challenge. I confess we haven’t yet formulated a global strategy for dealing with cybersecurity, but we do have in place a well-accepted and stringent strategy (and ethical code) for dealing with highly confidential and commercially sensitive data and the compliance requirements that come with it.  

3. Talent: Team building and attracting the right level of talent is always a challenge in any industry – more so in such a complex and highly specialized field as ours. As mentioned, we are not a company of computer geeks, you cannot computationally replicate biological processes without the right and broad mix of expertise (biology, informatics, software engineering, biotech/pharma). Our culture and geographic location have worked for us as we are able to attract world-class talent in all these fields.

Overall, we have reached a point in the progress of this field where pharma companies are starting to have a real and impactful experience from their "experiments" with machine learning. Our collaboration with Pfizer is a case in point, a significant and long-term collaboration, encompassing two of their biggest research units, that was the result of an extensive proof of concept process. This speaks volumes to the potential to improve on rates of success, de-risk clinical programs and enhance commercial opportunities by:

• Increasing the accuracy and speed of discovery (targets, compounds, and biomarkers), allowing for higher personalization, with accompanying earlier approvals in specific patient populations.
• Helping to reverse ongoing decline in R&D return on investment – a need that is driving an increased appetite for new technologies.
• Generating even greater value out of the increasing amount of molecular data which is now being generated in clinical trials, not just in research.

 In short, the opportunities are huge – both for us and the field as a whole.

LL: CytoReason recently announced a collaboration with Pfizer, could you tell us more about what this partnership will entail?

DH: This partnership is all about impacting R&D decision making throughout the pipeline, from discovery through target validation and on to clinical development and lifecycle management.

The collaboration will focus on immunology, immune-oncology, inflammation and dermatology indications and will bring data into the models that will further set them apart in terms of accuracy and strength.

The structure of the deal – involving significant technology access fees as well as success-based payments – is also interesting and is indicative of the broad spectrum of value that CytoReason brings. This not only includes potential IP but also invaluable biological understanding that may not represent as IP but can be critical in driving progress in the right direction, improving R&D productivity and reducing development risk.

David Harel was speaking to Laura Elizabeth Lansdowne and Ruairi J MacKenzie, Science Writers for Technology Networks.