NEW YORK – Precision oncology company Panakeia Technologies is expanding the number of indications for PANProfiler, its machine learning-based digital pathology platform.
A version specific to breast cancers, the PANProfiler Breast, received CE marking and UKCA (UK Conformity Assessed) certification late last year and the young English company is in early talks with the US Food and Drug Administration regarding possible US marketing approval.
"We're focusing our efforts on bringing our products to market — especially our first breast cancer product, as well as being able to support other cancers, into which we are looking to expand," said Panakeia CEO Pahini Pandya.
PANProfiler uses computer vision models to evaluate the status of a wide variety of biomarkers, including genetic mutations, transcriptomic and proteomic under- and over-expression, metabolomic pathways, and other established markers relevant for prognosis by analyzing digital images routinely collected from hematoxylin and eosin (H&E) stained tumor samples.
Jeremy Worley, a senior scientist at Bristol Myers Squibb who specializes in cancer systems biology, explained that pathological genetic mutations often lead to visible cellular changes.
When evaluating genomic alterations such as SNPs, he said, "most of them won't [associate with] an obvious morphological change. But if you're only looking at pathological SNPs, then you have a pretty small subset."
These, he said, "tend to be important in multiple ways. Even if it's a DNA damage protein, which you wouldn't expect to necessarily change morphology, it kind of does in downstream secondary ways."
Pandya pointed to the tumor microenvironment (TME) as an example of such a downstream morphological change. The TME has been attracting increasing research interest for its multifaceted roles in tumor evolution.
"Understanding how the tumor interacts with the surrounding stroma … the different extracellular matrix composition[s], and how the different types of stromal cells penetrate into cancers" are all visible at a high level and can inform many aspects of cancer, from diagnoses to tracking the effects of therapies, she said.
The relationship between biological signaling and morphological changes, said Pahini, makes the PANProfiler technique broadly applicable to a variety of cancer-linked markers.
Panakeia recently conducted a feasibility study, currently in preprint on bioRxiv, in which PANProfiler was applied to a wider range of cancers, as a step toward expanding the company's future commercial offerings beyond breast cancer.
The results showed that PANProfiler predicted biomarkers across 32 cancer types, including breast cancer, lung adenocarcinoma, colorectal cancer, and other major malignancies, with area under the curve measurements ranging from approximately 0.634 to 0.742. Markers included cancer-driving single nucleotide variations, transcriptomic and protein expression, and gene signatures. The strongest performance correlated with standard-of-care biomarkers.
While the study demonstrated that histo-morphological features were informative of cancer biomarkers, questions concerning specific mechanisms of biomarker detectability, such as how predictive patterns might be conserved across populations, will need to be better understood to move forward.
These upcoming validation studies, which are in various planning phases, will help determine when the next PANProfiler offering might make its way to market.
Despite a paucity of commercial digital pathology products like PANProfiler, Panakeia's offering will not be playing on an entirely open field. Several other companies have been developing their own products over the past few years, including Mindpeak, PathAI, Proscia, Ibex Medical Analytics, and Agendia.
Agendia's Digital MammaPrint, which the firm recently began offering through an early-access program in Brazil, functions similarly to PANProfiler, providing a diagnostic readout from H&E-stained slides that the firm considers equal to that of its 70-gene MammaPrint gene expression microarray.
Similarly, Panakeia is working with several "select" hospitals as part of an early-access program for PANProfiler. In addition to building the clinical evidence it will need to apply for regulatory approval, this may help establish a user base from which to grow.
Pandya commented that digital pathology is still in its early days, with potential customers actively evaluating a variety of tools and techniques.
"If you look at the digital pathology as a space on its own," she said, "there are very few hospitals that are digitized, and it's a new space where people are exploring what are other tools that they can use beyond digital pathology. For mainstream adoption, it's still a little while to go."
In general, digital pathology offers speed advantages over traditional immunohistochemistry methods, as well as greater ease of use. Digital MammaPrint, for instance, provides results within 24 hours, compared to four days for the microarray. PANProfiler, Pandya said, delivers biomarker results in "minutes."
Pandya and cofounder Pandu Raharja-Liu established Panakeia in 2018. The company currently employs just over a dozen people but Pandya said that this number is changing rapidly, as Panakeia is experiencing a growth phase.
"We are currently hiring for positions across research, engineering, and business operations," Pandya said, noting that these are primarily located in the UK.
The company held a seed funding round in 2019, and Pandya said the firm will release plans regarding more funding rounds soon. Current investors include Hoxton Ventures, Local Globe, and Entrepreneur First.