Abstract

Machine learning has seen some revolutionary
and remarkable developments over the last few years. Exciting breakthroughs have
been achieved in artificial neural networks, in particular around deep network
architectures (= deep learning), which
work based on raw-pixel image information and provide a classification
accuracy exceeding human expert judgement. Deep neural networks are now the
state-of-the-art machine learning models across a variety of areas, including image
processing and analysis and are widely implemented in academia and industry.

Genedata has developed the first
commercially available solution for deep learning-based High Content Image
analysis. Genedata Imagence^® allows for the application of deep networks to the
analysis of High Content Imaging, creating a workflow that cuts image analysis
time, increases data quality,
reproducibility of results and seamlessly integrates with Genedata Screener^®
for image data analysis. This deep learning approach outperforms conventional
approaches for feature extraction and phenotype classification.

Here we provide the evaluation of
a recent pilot of Genedata Imagence^® for High Content Analysis. We compare the Genedata
Imagence^® workflow with our conventional in-house High Content Imaging and
Analysis workflow through a series of assays that have been developed within
the target validation biology group at LifeArc. These include, immunocytochemistry
of shRNA knock-down clones, antibody internalisation assay, High Content cell
health assay and growth cone collapse assay​.
 The generation of training data and subsequent
training of the network was carried out for each assay and results of the data
analysis were compared to in-house conventional analysis. We discuss a few limitations
but show overall an excellent data quality produced by this novel deep learning
module for High Content Image analysis.