To get more from biological and clinical research, content is king. That’s why many scientists rely on high-content screening (HCS), which combines imaging and high-throughput techniques. Although HCS has been part of the pharmaceutical industry for some time, this technology promises to impact even more fields. The key to that expansion is making HCS easier to use.
In 2010, a team of scientists from the Centro Nacional de Investigaciones Oncologicas wrote an article called “High-content screening: seeing is believing”. That title concisely describes the key to HCS. As this group wrote: “HCS technology is integrated into all aspects of contemporary drug discovery, including primary compound screening, post-primary screening capable of supporting structure-activity relationships, and early evaluation of ADME (absorption, distribution, metabolism and excretion)/toxicity properties and complex multivariate drug profiling.”
With improvements in HCS technology, it can be used in even more ways. Also, some HCS platforms reveal information that even scientists can’t find on their own.
Simplifying implementation
At BioTek Instruments, the “current focus is to provide customers with simple, yet powerful, tools to seamlessly process and analyze large quantities of images captured by our automated microscopy systems,” according to imaging and microscopy specialist Chris Laucius.
“These systems analyze a collection of metrics on individual cells to characterize populations. As Laucius notes: “Gen5, BioTek’s imager software, is equipped to handle these requirements by automatically identifying and measuring multiple metrics of individual cells.” Recently, the company expanded these metrics with Gen5’s Spot Counting Module and enhanced data visualization tools, including histograms and scatterplots.
When asked about the key features of this advance, Laucius says, “Gen5 has an intuitive user interface that guides the user through the entire imaging process.” This covers the entire HCS workflow, from image acquisition, processing, and analysis to producing results.
Laucius and his colleagues hope to expand HCS applications. “Our goal is to make the complicated realm of high-content image analysis more accessible to life scientists,” Laucius adds. “As image-based methods are becoming indispensable for many biological applications, having tools available that do not require extensive expertise in the field of image analysis is critical.”
Developing data
Data make up the basis of an HCS run, and dealing with it all takes time and usually expertise. One key, though, to expanding the use of HCS is making it ever easier for users to handle the data.
That’s why Genedata first developed its Genedata Screener software. “This is an enterprise-level data analytical solution for many different plate-based instruments, which is tuned for very efficient and scalable processing of numerical results, such as traditional image analysis performed by high-content instruments and their software from other companies,” says Stephan Steigele, head of science at Genedata.
Going a step further, the company recently released its Genedata Imagence, which performs image analysis and delivers the numerical results to Screener. “Imagence is a deep learning–based software,” Steigele explains. This software removes many of the time-consuming human interventions required in the past.
“In classical computer-vision analysis, a human must handcraft the features,” says Steigele. That is, there is often some manual adjusting required to analyze aspects of an image, such as cell size or the intensity of a labeled protein of interest in the cytoplasm. And effects between batches can dramatically differ; for example, the amount of a labeled protein expressed could make the intensity too high or too low in an image, and the user would need to adjust for that. “These are all aspects that a human has to think about,” Steigele explains.
Those are exactly the kind of adjustments—manual tweaks—that Genedata Imagence does not require. “This software enables a very, very efficient way to generate training data,” Steigele explains. So, instead of needing one or more imaging experts, just a single biologist can use this software. “Here, you have an app where you can intuitively explore phenotypic space, and then triage cells in different classes, like ‘bright signal’ or ‘not so bright’,” Steigele says.
The key to Genedata Imagence is that the deep-learning based procedure autonomously learns which features to extract. Moreover, the software performs this process in a fast and unbiased way. The same cannot be said of doing it manually. “A classical approach to finding a pharmacological endpoint from image analysis can take weeks, because humans often work in teams with computer scientists and image-analysis experts,” Steigele notes. “With our image solution that includes no human reasoning, biologists can get to that endpoint in just four to six hours.”
Rather than just claiming better performance, Genedata proved that its software improves HCS results. Working with AstraZeneca and Amgen, scientists at Genedata compared the pharmaceutical companies’ legacy analysis with the results from Genedata Imagence. “The deep-learning results often exceeded the results that a human could produce,” Steigele says.
Breaking into biology
Some of the most intriguing uses of HCS ahead could impact basic biology. That application is already underway, and likely to spread out across various biological disciplines.
In 2016, Brenda Andrews from The Donnelly Centre at the University of Toronto and her colleagues wrote that HCS “has been used to address diverse biological questions and identify a plethora of quantitative phenotypes of varying complexity in numerous different model systems.” These scientists describe additional applications, including identifying genes and characterizing their interactions. As this group noted: “The growth of HCS has been driven by advances in biology and chemistry, as well as mechanization and computation.”
Andrews and her co-authors describe many interesting applications, some of which might not be expected. For example, they discuss using fluorescently labeled versions of the huntingtin protein in fruit flies to study the pathology of Huntington disease.
Despite all of the benefits of HCS, Andrews and her colleagues point out some remaining questions. As an example, they ask: “How do we integrate HCS data with other large-scale data sets to get a complete picture of the cell?” And they wonder: “What type of coherent strategy for data presentation and accessibility can we generate to unify reporting standards and make data from different sources directly comparable?”
That’s the thing about high content: There’s a lot of information to consider, and that creates questions about how it can be used. Even as questions remain, it’s clear that HCS will continue to be used in exploring basic and clinical questions.