NEW YORK – Pangea Biomed on Thursday said it would use $7 million in seed funding to use its transcriptomics-based ENLIGHT platform to get more cancer patients on precision oncology treatments.
Tel Aviv, Israel-based Pangea's Expression Networks for highLIGHting Tumor vulnerabilities, or ENLIGHT, uses a combined DNA-RNA machine learning approach to predicting whether cancer patients will respond to specific treatments.
"The unique thing about Pangea's approach is that it entails looking at not only the DNA of the tumor, but also at the RNA," said CEO Tuvik Beker. Although there are routinely used RNA-based tests from firms like Tempus or Caris Life Sciences, Beker explained that these tests are primarily used to identify gene fusions associated with treatment response.
"Other than that, the vast amount of information that's in the RNA-seq is largely neglected," he said. "Pangea's approach shifts the floodlight from the DNA, where everyone is looking for actionable mutations, and onto the RNA expression data analysis."
Pangea's ENLIGHT test provides patients a matching score indicative of the treatments they're likely to respond to. The firm is betting that the ENLIGHT Matching Score, or EMS, can determine whether a patient will respond to an immunotherapy or targeted therapy, even if they don't have obvious targetable genomic alterations or elevated expression of single genes.
The test requires that patients' tumors are first assessed using whole-exome and whole-transcriptome sequencing. The firm then combines patients' sequencing data with information on network gene effects using unsupervised machine learning. The algorithm draws on information about activation patterns in networks of genes from clinical, in vitro, in vivo, and phylogenetic data available through publicly available sources and through collaborations with academic research groups.
The firm is currently offering its ENLIGHT test on a research-use only basis to evaluate patients' likelihood of responding to a list of over 100 therapies. The test not only matches patients to drugs they might respond to but also ranks them based on whether the recommendation is for an on-label, off-label, or experimental therapy.
"Our algorithms are mechanism-agnostic," Beker said. "They work on checkpoint blockers and other monoclonal antibodies as well as small molecule inhibitors."
Importantly, while one of Pangea's key selling points is that ENLIGHT could, in theory, make precision oncology accessible to more cancer patients, the test's value is directly tied to the quality of the available genome-level data for each treatment it considers.
"We're only as good as the data regarding the drug," Beker said. "What we rely on is a good characterization of the drug's targets."
For this reason, currently the ENLIGHT test is particularly well suited to determining whether patients will respond to monoclonal antibodies, since the targets of these drugs are very well characterized. "They target some very specific proteins, and there are practically no off-target effects," Beker said.
For small molecules and some tyrosine kinase inhibitors, he continued, the "mileage may vary," since the number of off-target effects may be greater and "their prediction is a little bit more noisy."
"But generally, we've proven that in all of these classes of targeted therapies, we have good predictive capabilities," he said.
Currently, the approach is based on understanding functional interactions between individual genes. Reducing the resolution of the algorithm to look at interactions between pathways could make the test applicable to chemo, too. According to Beker, doing so is an "active area of research" for Pangea.
Last April, researchers described a similar approach to ENLIGHT, via a precision oncology framework dubbed synthetic lethality and rescue-mediated precision oncology via the transcriptome, or SELECT, in Cell. In that paper, they showed that by combining synthetic lethality and synthetic rescue data — information on which gene pairs work in tandem to promote or prevent cell death — with tumor transcriptome sequencing, they could correctly predict patients' responses in 28 out of 35 targeted therapy and immunotherapy drug trials. That study built on a prior prospective umbrella trial dubbed WINTHER in 2018, which was the first clinical trial to prospectively match patients to treatment with transcriptomic data.
However, as Beker explained, the approach used in ENLIGHT differs from the one employed in the SELECT study in one key aspect. With SELECT, researchers considered pan-cancer functional gene interactions two ways: by looking at pairs of genes with synthetic lethality and pairs of genes with synthetic rescue. Like ENLIGHT, SELECT gleaned these gene interaction sets from available cell-line and patient genome databases — namely, from the Cancer Genome Atlas program — and used these sets as a backdrop against which to analyze patient RNA-seq data for potential drug response. Ultimately, within SELECT, patients' "scores" were determined according to the percentage of the identified gene interaction sets that were activated in their sample.
ENLIGHT, meanwhile, looks at large "social graphs" of functional gene interactions rather than gene pairs implicated in synthetic lethality or synthetic rescue. "We're looking for traces or evidence of functional relationships in big datasets of cancer patients and normal tissues," Beker said. "We've got a very strong engine that ingests all of that big data and the output is those maps, or social graphs as we like to call them, saying 'who's friends with whom,' or which genes are helping other genes when they're hit and which genes are actually detrimental to other genes."
Armed with those network maps, the platform's ability to predict drug response is not reliant only on information about a therapy's specific target or indication, and therein lies its advantage.
"What we do at the predictive stage is take the patient's very unique tumor, sequence it, and then for any drug that you want to screen, we say, 'Ok, what is that drug's target?' and we widen the aperture. Instead of looking just at the target gene, we look at the whole network of gene interactions surrounding it."
To date, Pangea has applied its approach retrospectively to 21 patient cohorts beyond those considered in SELECT. These additional cohorts spanned 11 cancer indications and 15 therapies, and investigators were blinded to the clinical outcomes until the end of the study.
The analysis demonstrated ENLIGHT's ability to generate predictions for all of the cohorts. These predictions were associated with improved response to treatment in all but two cohorts, improving upon the efficacy seen with SELECT. Furthermore, ENLIGHT performed as well as other supervised predictors already developed for predicting patient responses to these drugs.
Pangea hasn't published the findings but plans to do so in a peer-reviewed journal in the next few weeks.
The next step for ENLIGHT, Beker said, is to prospectively trial the test's performance in clinical trials that the firm expects to announce during the third quarter. And while he was unwilling to share specifics about these studies, he was quick to point out that the firm has already been offering ENLIGHT to oncologists — mostly doctors practicing in Pangea's base country Israel — for research use only.
"Any patient coming in with DNA or RNA sequencing results can get an ENLIGHT test for free if their oncologist thinks it might be of interest in a particular case," he said, sharing an anecdote about a patient in Israel with a rare, advanced-stage cancer who is still in remission because ENLIGHT matched her to a therapy that was "completely contrary to standard biomarkers." Beker withheld details about the type of cancer this patient had or the ENLIGHT-predicted treatment, because Pangea will submit the case report for publication soon.
Going forward, Pangea expects oncologists and drugmakers to be customers for ENLIGHT. After the test is ready for full commercialization, oncologists will be able order ENLIGHT for patients just as they would an NGS cancer profiling test from Tempus or Caris Life Sciences. "We are collaborating with a large diagnostics lab to bring ENLIGHT tests to the market in a bundled offering on top of an existing molecular profiling test," Beker said, but declined to name the Dx partner.
Pharmaceutical companies, meanwhile, can use the test to refine their clinical trial inclusion and exclusion criteria. Beker said the firm is already collaborating with several pharma companies "to more efficiently direct their drug development by looking at the best patient sub-populations." The pharma partnerships aren't yet public.
Even though Pangea is hoping to use ENLIGHT to make precision oncology available to more patients, it may not be completely immune to access barriers endemic in the field because the test still relies on sequencing patients' tumors. Pangea outsources the sequencing portion of the test, and it can take several weeks to receive the results. This is time that many advanced cancer patients may not have to spare before starting treatment. Beker acknowledged this but added that Pangea is doing its best to ensure that the additional analysis portion with its machine learning algorithm takes fewer than 48 additional hours because "time is of the essence."
Additionally, Pangea has so far validated ENLIGHT only on formalin fixed paraffin embedded tumor samples, or bone marrow samples for hematological cancers. Procuring tissue from heavily pretreated, advanced cancer patients such as those with lung cancer can raise safety concerns. Pangea hopes to get around this challenge by validating ENLIGHT with liquid biopsies in the future.
"We believe that this will really revolutionize cancer care once we can do that [blood-based testing]," Beker said. "I really hope we can make precision oncology available and beneficial to many more patients."