NEW YORK (GenomeWeb) – Fresh off a $2.8 million seed funding round, Swiss startup Scailyte aims to use its automated single-cell artificial intelligence-based analysis platform to help researchers discover complex disease patterns to identify diseases in their early stages.
The Sursee, Switzerland-based firm plans to launch the platform, called ScaiVision, for research-use-only applications in the second quarter of this year.
Scailyte began as a collaboration between CEO and Cofounder Peter Nestorov and Scailyte Cofounder and Scientific Advisor Manfred Claassen, who worked in the single-cell biology research space. They saw commercial potential for a niche data generation technology, and as well as a need for extracting clearly defined results from single cell data. Partnering with two software engineers, the team built a system that identified patterns associated with disease status.
ScaiVision consists of several modules, such as data visualization, data handling, and data analysis. One of the data analysis modules uses an artificial intelligence approach called CellCnn — developed in Claassen's lab — which uses convolutional neural networks to find condition-associated cell subsets from high-dimensional single-cell data. The algorithm then identifies certain conditions — such as disease or drug response —linked to cell populations.
According to Claassen, the analysis enables researchers to find differences across single-cell datasets and involves "hypothesis testing to evaluate whether these differences are statistically significant (and not mere random fluctuations)."
Claassen said that ScaiVision allows users to visualize differences as well as assess their statistical significance using "hypothesis testing," which is a method to rule out the possibility that observed differences are occurring because of random fluctuations. By eliminating the need for a tedious and subjective manual visual inspection, Claassen highlighted that ScaiVision performs the step within a few minutes, rather than several weeks.
After researchers extract the target cells and build protein-based single-cell data, they can upload the information onto Scailyte's website using an online portal. The user can select up to around 50 parameters before running the software to generate a curated report to identify clinically relevant patterns, which highlight phenotype-related cells.
According to Nestorov, ScaiVision can generate a customized report for users within one to two days.
While ScaiVision currently supports analysis of data derived from protein expressions through mass cytometry and flow cytometry, Nestorov noted that the firm plans to expand to single cell RNA-seq data in the future; however, the tool is neither tested nor optimized for the data at this time.
According to Scailyte Cofounder and Director of Data Analytics Daniel Sonnleithner, the report's dataset will contain the intensity levels of each parameter for each single cell of the sample. Metadata, such as when the firm ran the initial experiment, which instrument the team used, is also included.
"The idea of the company is to cover all types of single-cell data as well as to stay platform agnostic," Nestorov said. The firm has begun an active clinical study with an academic collaborator, but Nestorov declined to say more about the partnership.
However, Sonnleithner noted that Scailyte is also running an in-house project that makes sure that the results of the AI module match users' expectations. The team is using data manually analyzed by previous research groups to check if the software is producing similar results in an automated fashion in a shorter time span.
Limitations
Sonnleithner acknowledged that the team has encountered several challenges while improving the AI algorithm. Not only did the module need to work with researchers who understood how the technology works, but the firm's engineers also had to ensure that it was robust enough to perform in situations where it ran 24 hours a day, 7 days a week to digest data from different labs.
In addition, Sonnleithner noted that the firm also needed to develop the system in order to handle data from several thousand customers, rather than just a few users.
"We didn't improve the artificial intelligence algorithm with respect to its performance, however we made it production ready," Sonnleithner explained. "Furthermore, the great majority of the work went actually into the surrounding systems that allow users to work with such an artificial intelligence module."
Nestorov envisions the technology being used in the research space to analyze single cell data. Within the clinical space, he believes that single cell data contains information that will help physicians "identify rare cells or very complex patterns, which would lead to much more sensitive diagnostics compared to the state of standard care."
In addition, Nestorov said that the Scailyte will also eventually move toward developing diagnostics for early and more precise detection of disease biomarkers.
"We're now looking to identify which sector really have an urgent medical need and would most need an application for their market," Nestorov said.
Scailyte also offers ScaiBmD service, partnering with pharmaceutical and biotechnology companies to provide its technology for biomarker discovery for early and precise disease detection. Sonnleithner said that while the firm was developing the ScaiVision software, it identified another need in the single-cell technology market.
"We understood that a lot of our potential customers don't have the required know-how in house to design and run experiments with this new technology," Sonnleithner explained. "Therefore, we are offering our expertise in the field as a service to externals," which should "help to lower the entry barrier to this exciting technology and helps us better to understand the exact needs of the market."
Scailyte will have to contend with existing companies and research groups offering their own single-cell analysis methods.
Fluidigm currently offers its Hyperion Imaging Mass Cytometry platform, which the firm said allows users to add sample spatial and structural data to single-cell protein data generated by its mass cytometry systems.
Over the past year, 10XGenomics commercially launched multiple single-cell analysis products for its Chromium controller platform as it plans to expand commercially in 2019. The firm also made several acquisitions in 2018, including Spatial Transcriptomes and Epinomics.
Nestorov argued that ScaiVision's unique selling point is it has automated the hypothesis testing portion of the analysis. By offering a cloud-based platform for researchers, Nestorov believes that the step helps researchers visualize and explore the data in a simplified report.
At the same time, Sonnleithner noted that ScaiVision relies on already pre-processed data, rather than raw sensor data. The tool is therefore complementary to software packages provided by instrument manufactures. ScaiVision does not support imaging mass cytometry data that has not been pre-processed, for instance.
Scailyte plans to launch the tool for research-use-only applications in Q2, targeting European research groups looking to rapidly manipulate single-cell data.
The company will eventually accept requests from research groups worldwide.
"We plan to address the technical side to properly identify cell types, because when you have single-cell data, people who would like to know what types of cells really are in there with an automated fashion," Nestorov explained. "A biologist needs to see behind the computer and look at the expression of different markers on some of these cells, and so we're looking into automating cell type annotation."
According to Nestorov, Scailyte is currently finalizing the pricing for its product, but he noted that it will depend on the type of user — for instance, academic or industrial, or from a research group or a core facility.
"There are a lot of players out there coming out with new platforms, new chemistry, and data types every month," Nestorov said. "So, if you stick to one type, you may become obsolete pretty fast, [but] the beauty of this algorithm is that it can be applied to multiple types after data has been processed to a certain extent."