Deepcell, a pioneer in AI-powered single-cell analysis to fuel deep biological discoveries, has made three datasets available online for academics to examine single high-dimensional morphology data. Datasets were generated using DeepCell’s high-throughput platform, which includes image processing and classification equipment, artificial intelligence algorithms and software.
Researchers can easily generate high-dimensional readouts of known and novel morphological features of unlabeled cells using the AI Human Foundation Model (HFM) model, which has been trained on millions of cell images. This is done using an unrestricted hypothesis technique. This data release will also include a software package that allows for the generation of single cell sorting and the identification of morphologically relevant cell clusters for viable cell sorting.
This initial round of data releases demonstrates how deep cell technology can investigate specific cell populations of interest that are difficult to identify with molecular markers and characterize different cell types in a heterogeneous sample in a marker-free manner.
observation of morphology
There are three human cancer datasets to explore, all of which contain samples purchased from reputable vendors, including Discovery Life Sciences, Cell Biologics, and BioIVT. According to Giovanna Prout, vice president of marketing, in a message to BioIT World.
In an initial data set, 28 samples of human melanoma cell lines and real tumor samples were used using the DeepCell platform to identify tumor, immune and stromal cell populations in a marker-free manner using only morphology. One group, according to Prout, uses about 28,000 cells via the UMAP Projection Deep Cell software package for easy visualization.
To gain more clarity on this morphologically unique subpopulation, melanoma tumor cell population data from this dataset were selected in the DeepCell software package and a custom UMAP was used to generate another dataset of approx. 350,000 cells. According to Prout, the UMAP projection uses an array of approximately 18,500 cells via the DeepCell software package for easy visualization. According to the researcher, it “elucidates variations within cells based on minimal morphological variations, such as pigmentation, which can be difficult to perceive with conventional methods”.
In the final data set, a series of samples of human differentiated tumor cells (DTC) were used to study the morphological diversity of immune cell populations in the lung tumor microenvironment using deep cell label-free technology. The third data set consists of 11 samples that were performed on the DeepCell platform, resulting in a total of approximately 1.1 million cells. According to Prout, UMAP Projection uses an array of approximately 17,000 cells via the DeepCell software package for easy visualization. “The morphological variation is surprisingly high and complex, even within a single sample. No single set of features or classes can adequately describe this amount of data. Basic models and self-supervised learning Our method provides, in our opinion, an unmatched advantage when studying.
According to DeepCell, the ability for the scientific community to visualize high-dimensional single-cell morphology data generated on the DeepCell platform is one of the goals of the data release. To understand the kind of analysis that DeepCell produces through its platform and to increase researchers’ curiosity about the possible applications of this data in the context of biological paradigms and scenarios, Prout emphasized that researchers need to create datasets. She hopes that the data sets will generate new ideas about how the technology can be used for new research.
