(1210-C) Advances in multimodal image data management in OMERO Plus: connecting tissues, cells, sequences, and features
Tuesday, February 6, 2024
12:00 PM – 1:00 PM EST
Location: Exhibit Halls AB
Abstract: OMERO Plus is an image data management platform designed for storage and analysis of bioimaging data and metadata. Backed by the Bio-Formats library, OMERO Plus can natively store and retrieve image formats arising from a wide variety of microscopy hardware, from high content screening (HCS) systems to spatial transcriptomics platforms. Metadata can range from simple annotation tags to rich tables describing per-object (cells, nuclei, spots, and other structures) features. Together, this means that large and complex image datasets and their associated sequencing, segmentation, or other analytical results can be both stored and retrieved remotely.
The utility of this system is enhanced by closer integration with existing data science tools for processing stored data and optimized performance of remote data retrieval.
To provide a more convenient method for working with tabular data in OMERO Plus (OMERO.tables) from Python environments, Glencoe Software released the open source omero2pandas package (https://pypi.org/project/omero2pandas/). This package allows users to download (complete or partial via custom queries) and upload OMERO.tables data as pandas dataframes, with additional integration aimed towards Jupyter Notebooks. This provides access to the full suite of Python data science and machine learning packages when working with OMERO data. Omero2pandas is another step towards making OMERO Plus the data engine of choice for data analytics and AI in bioimaging.
We evaluated the performance and user experience of the above data retrieval operations for three use cases: cell segmentation of highly multiplexed fluorescence whole slide images, cell segmentation of Cell Painting HCS datasets, and spatial transcriptomics datasets. Omero2pandas and OMERO.tables enable faster retrieval of tabular data, in some cases by orders of magnitude, with the addition of support for table compression and table querying for selective data retrieval. We also demonstrate the usability of this tool for dimensionality reduction on a table detailing millions of objects with hundreds of features, including the visualization of these results in open source and commercial tools.