Genedata's 'No BS AI'
By Stephan Steigele, Matthias Fassler, Thomas Hasaka, Stephan Heyse
July 22, 2019 | In April, we announced the winners of the Bio-IT World Best of Show Awards during the Bio-IT World Conference & Expo. These awards are given with the goal of recognizing the best of the innovative product solutions for the life sciences industry on display at the conference. We wanted to highlight these products as they measurably improve workflow and capacity, enabling better research. – The Editors
Given the impressive number of AI applications currently available in the life sciences, the Genedata team was most honored to have Bio-IT World judges award Genedata Imagence with the Bio-IT World Best of Show Award in 2019. In particular, we were gratified for winning the "No BS AI" category, as our solution represents a true practical AI- and customer-driven application, which uses the full power of deep learning to enable workflows that will dramatically change phenotypic screening in early drug discovery and beyond.
What makes this solution so unique and why did we win the "No BS" award? Understanding the judge's decision requires an understanding of our design principles: We develop enterprise software for practical applications in life science R&D together with the scientists at our industry partners. Here, we developed software for automated analysis of massive amounts of image data, generated by highly automated microscopes in small or large molecule screening campaigns (often involving testing of many 100,000s of compounds). Genedata Imagence empowers a single assay biologist to simply run this massive data analysis in self-service mode. This represents a sea change in their daily work. Consider this: In a classical high-content screening (HCS) workflow, establishing the analysis procedure and adapting the assay to it is labor- and time-intensive. It requires scarce experts who oftentimes work on teams with different roles and people and involves several handovers and iterations. This arduous process demands stringent coordination and quality control to guarantee robust working assays. With our deep learning-based HCS workflow, the same results can be generated by a single scientist who generates and curates the training data using just the HCS images as reference, and let Imagence do the rest. Scientists do not have to worry about perfecting the assay to a particular analysis workflow. This image curation is the only hands-on step, and it is fast: our solution reduces the time for image data analysis from weeks to just a few hours—and that's from raw data to pharmacological quality results.
The Need for Efficient Phenotypic Image Data Analysis is on the Rise
The drive towards higher biological relevance of in-vitro assays has led to an increase in phenotypic HCS, and so too increases the need for more efficient image data analysis solutions. Differentiating cellular phenotypes is key to obtaining reliable pharmacological information: 1) as stable endpoints for primary drug response; 2) for assessment of toxicity and safety-relevant effects; 3) for the discovery of previously unexpected drug effects. Here's where our AI-driven solution makes the difference: it enables a highly efficient workflow for phenotype curation of new or unknown phenotypes. Typically, a phenotype is defined by an agreement between experts on its defining visual characteristics. With Genedata Imagence, however, scientists quickly gain unique, unbiased and reliable access to this sort of expert phenotype identification and classification—without having to set up any image analysis or even needing to know how to do it (image analysis set up means handcrafting of many hundred—to thousands of features taken from the pixel information in high-content images).
To date, in more than 15 industry projects, Genedata Imagence has delivered results that correlate well with those generated by the classical HCS analysis approach, albeit in a much faster and simpler fashion. This is due in large part to close and successful collaborations with our industry partners who have worked with us in developing this solution. For example, our work with AstraZeneca, which garnered the 2018 Bio-IT World Best Practice Award, enabled us to tailor the underlying AI technology to produce the most meaningful pharmacologically relevant results in which compound potency values had been stressed during very critical comparisons to gold standard analyses at major pharma companies. This collaboration helped us to develop a robust deep learning-based software that eliminates the need for re-engineering across very different assay types and settings while enabling a workflow that is as easy to use as an iPhone.
The Bio-IT World judges have awarded Imagence as being "No BS AI"—and indeed this has been proven by the analysis of industry-size screening data sets by our customers. The results have been robust and comparable on all levels meeting the gold standard analysis demanded by many pharma companies for classification accuracy, plate QC parameters, and compound potency. While it can be argued that comparable or even enhanced result quality can be achieved with classical HCS image analysis, Imagence produces robust AI-based results in a fraction of the time typically devoted to the classical gold standard analysis and delivers on the complex outcomes typical for phenotypic assays in today's pharma research.
We thank the Bio-IT World judges for recognizing the value of this solution. Moreover, the Best of Show Award further motivates our team to continue to push the envelope in developing No BS AI applications that fuel innovative drug discovery.
Dr. Thomas Hasaka is a scientific account manager at Genedata supporting both Genedata Imagence and Genedata Screener. Formerly with The Broad Institute, he has expertise in High Content and High Throughput Screening, microscopy, and laboratory automation. Thomas earned his PhD in biology from Temple University. He can be reached at firstname.lastname@example.org.
Dr. Stephan Heyse is head of the Genedata Screener business unit where he leads ongoing development and support for Genedata Imagence and Genedata Screener. Previously he led a high-throughput screening unit at Bayer and managed the founding industry partnerships for Genedata Screener. Stephan earned his PhD in biophysics at École polytechnique fédérale de Lausanne. He can be reached at Stephan.Heyse@genedata.com.
Dr. Matthias Fassler is scientific account manager for Genedata Imagence and Genedata Screener. Passionate about imaging-based approaches, his work centers on bridging the gap between biologists and software developers. He earned his PhD in cell biology and neurobiology from Friedrich Schiller University Jena. He can be reached at email@example.com.
Dr. Stephan Steigele is head of science at Genedata. Expert in life science and pharma data analysis and R&D, he leads development of Genedata Imagence while focusing on areas such as AI, automated screening technologies, and multivariate data analysis. He earned his PhD in biology and computer science from the University of Leipzig. He can be reached at firstname.lastname@example.org.