Skip to main content
Premium Trial:

Request an Annual Quote

NanoCellect Biomedical to Launch New Imaging-Based Cell Sorter With Potential for AI Plug-Ins


NEW YORK – NanoCellect Biomedical is planning to launch an upgrade to its Wolf G2 cell sorter this quarter, with commercial installations coming by the end of the year.

The new instrument, the Verlo Image-Guided Cell Sorter, adds 2D imaging capabilities on top of the gentle handling its other model is known for. This will enable several key applications, including identifying and sorting cells with functional cell-cell interactions, subcellular localization of markers, and label-free cell analysis to guide sorting decisions. The instrument offers throughput of thousands to millions of cells per run and analysis of 2,000 events per second or a sorting rate of 200 events per second.

The San Diego-based firm has launched an early-access program for Verlo and is targeting collaborations, especially in those three application areas.

"Verlo will take some Wolf users but open up many new users as well," said Paul DiGregorio, VP of commercial at NanoCellect, noting that the company will continue to support its Wolf product line. For customers focused on fluorescent markers, imaging may not add as much value, and they may prefer to stick with the Wolf over the more expensive Verlo, he added.

DiGregorio declined to disclose the pricing of Verlo, but said the firm is offering incentives, potentially including special pricing, to "scientific visionaries who want to gain early access."

While Verlo won't ship with artificial intelligence-based sorting algorithms, such decision support will be available through third-party software, and the firm is working on embedding those capabilities to the platform at a later date. "Now, [AI] is a feature requirement more than anything," DiGregorio said. "We're making sure we're compatible."

Already, one of the firm's cofounders has developed AI models that can use Verlo's imaging scheme to make sorting decisions, possibly even without the use of training data sets.

With Verlo, NanoCellect joins several other companies offering imaging-based flow cytometry or cell sorting technology. BD's FACSDiscover S8 launched last year, featuring CellView Image Technology.

And other cell sorting startups with an eye toward AI integration, such as Stanford University spinout Deepcell and Japan's ThinkCyte, have developed and launched platforms. In January, Deepcell partnered with Nvidia to develop generative AI technologies for cell analysis. ThinkCyte launched its VisionSort instrument last year.

ThinkCyte offers 3D analysis of cells, a direction that NanoCellect could be heading.

In a presentation at last week's Molecular Medicine Tri-Conference in San Diego, Yu-Hwa Lo, a NanoCellect cofounder and researcher at the University of California, San Diego, presented data form his lab's work on 3D imaging-based cell sorting.

DiGregorio stressed that Verlo is a 2D imaging system, though it has a collaboration with Lo on 3D imaging and shares IP on the technology. The firm's market research shows that, for now, 2D is a higher priority, he said. "There are some scenarios where 3D would be enabling," he said. "It's an evolving application area. The market feedback is that 2D image-guided sorting would have a stronger fit."

Lo's 3D system generates tenfold more data than the 2D system — on the order of terabytes per run, so the use of AI is a necessity, he said, in addition to improving the performance of AI-based sorting.

In an experiment to classify HeLa and MCF-7 cells, the 3D system was able to classify cells with around 98 to 99 percent accuracy, up from around 80 to 90 percent for the 2D system. And it was able to perform label-free sorting of human whole blood cells such as monocytes, lymphocytes, and granulocytes "with almost 100 percent accuracy."

This was with only one onboard CPU and one graphics processing unit (GPU). "In two milliseconds, more than 99 percent [of cells] can be classified with a 95 percent confidence interval," Lo said. "We can easily upgrade the AI hardware with GPUs and then we could achieve less than 100 microseconds."

Applying an unsupervised learning method, where the algorithm does not need to be trained with application-specific data, also led to positive results. In a proof-of-concept experiment, his lab asked the AI model to identify and sort leukemia cells from regular cells without any labeling, something that expert humans could not do. It identified the leukemia cells with 99.9 percent accuracy and regular cells with 95.4 percent accuracy.

AI has also shown promise in identifying drug-treated or thermally stressed cells this way, or even to predict the top-performing Chinese hamster ovary cells in terms of protein generation. "This could open doors for a lot of new applications," Lo said.