Lab automation has moved beyond liquid handling robotics to include a variety of technologies that help lab technicians do more, and do so more effectively. Smart instruments are behind this, incorporating multiple features, storing data in the cloud, and sometimes sharing that data with other instruments downstream. Often, cloud-based analysis is performed automatically so that lab professionals see results and their interpretation simultaneously.
These capabilities are iterative. As labs accept cloud computing, the Internet of Things (IoT) becomes more useful and more prevalent. As more data is collected and stored in widely-accessible compute clouds, artificial intelligence (AI) and its subset, machine learning (ML), become increasingly practical. In fact, AI and ML are leading trends within life science labs today.
“The advantage of automation is that instruments are more powerful, can aggregate data more quickly, and can find analytical insights that otherwise wouldn’t be found,” says Henrik Gehrmann, vice president of engineering at Clear Labs. “Using such devices—sometimes even utilizing the IoT, which places sensors at the edge of your network—as part of a bigger solution really empowers labs to solve problems they couldn’t otherwise solve.”
For example, Gehrmann says, “That could mean taking output from one device as input for the next…like using the lab information management system (LIMS) to track patient data, pool that data in the cloud, and use machine learning to detect patterns to create something that gives insights (often in a few hours) that otherwise would take weeks of manual work.”
The case for automation and AI
AI is becoming an important element in laboratory analysis, yet “there’s a misconception around automation and AI,” notes Richard Lee, director of core technology and capabilities at ACD/Labs. “You don’t necessarily have to have both. They aren’t dependent on each other.”
Automation is for efficiency. It’s used to streamline processes or to transfer data and make it more widely available, either from multiple locations or devices for your own team or to collaborating teams. In contrast, AI uses the data collected from instruments to identify patterns, such as analyzing video to understand exactly how flies beat their wings during flight or to count neuromuscular junctures in mice. ML takes this a step further by making predictions (such as predicting a disease prognosis based upon certain conditions). Singly or combined, these technologies can eliminate a lot of tedious, manual work.
Many lab managers think AI has a high learning curve, “so there’s inertia, particularly for small and medium labs,” says Mark Fasciano, co-founder and CEO of Rover Labs. Lab managers also have to gauge whether the cost and time savings are enough to justify adding AI to the lab. “For many, the answer is ‘maybe not,’” he admits.
“The advantage of automation is that instruments are more powerful, can aggregate data more quickly, and can find analytical insights that otherwise wouldn’t be found.”
Yet, automation and AI can add a lot to labs that need it. The development of fully-automated next-generation sequencing during the COVID-19 pandemic is one example. Clear Labs’ whole genome sequencing instrument, for example, combines sequencing, robotics, and cloud-based analytics. It initially was developed for food safety, to screen for salmonella and listeria in poultry, but the lab redirected the technology for diagnostics when the pandemic hit. With runs of 32 to 64 genomes, Gehrmann says this is a good fit for small- to medium-sized public health labs. Automating genome sequencing streamlined the process, and sending the testing results to the cloud made it easy for labs to notify patients the next day if they tested positive for COVID-19, and to track known and emerging viral variants quickly and thus better contain transmission of the virus.
In contrast, “Traditional, legacy, next-generation sequencing was manual and tedious. Public health labs needed two to three days to do the prep work and start the sequencing and another four to 10 days to identify the variants,” says Jeff Field, chief commercial officer, Clear Labs.
Another AI-enabled diagnostic (developed by Texas A&M System and Worlds Inc. under the Rapid Acceleration of Diagnostics-Radical initiative at the National Institutes of Health) uses AI to analyze breath tests, Fasciano says. A patient breathes into a bag, the contents are analyzed using mass spectrometry, and AI interprets the results in less than a minute with PCR-like accuracy. “It’s like a version of computer vision. To a human, the data looks like static, so the only way to interpret it is with AI,” Fasciano says.
Start with the cloud
Even though the cloud has been used in business for many years, it’s only recently gaining popularity for scientific applications, Lee says. Acceptance was accelerated by the work-from-home mandates during the COVID-19 pandemic that closed many labs and slowed many projects.
To integrate AI into your existing lab operations, start by adding cloud computing. This is the foundation that transforms data from instrument-based silos into aggregated datasets that can be accessed, searched, and analyzed by far-flung teams that need to access both historic and current datasets.
Even if the data stays entirely within your lab, you can benefit by enhanced data transfer capabilities. “Scientific data files can be very large,” Lee says. “For example, a single LC-MS data set from a high-resolution instrument can be up to about 2GB from just one run. Transferring one data set is burdensome, and transferring 10 to 100 runs per day is quite a bit.” Therefore, storing the data in the cloud as it is created eliminates the need for separate transfers. It also enables access to approved users on- and off-premises, as well as the ability to store enormous quantities of structured and unstructured data.
Software-as-a-Service (SaaS) applications are another option, Gehrmann says. “SaaS is an online cloud-based service that puts computational power at your fingertips for a specific application. It can scale on demand. It’s more beneficial for small- and medium-sized labs than for large ones, although the big labs also are moving to the cloud.”
Once data is easily accessible, ML may be beneficial. For ML to be most useful, lab managers need to work with their staff to organize data in a specific format and to clean and normalize it, so the data can be searched and deliver accurate, comprehensive results.
Many of the data formats (such as from the mass spec machine) may be proprietary, and even simple things like dates are described differently. Bringing all the data together into a single, searchable platform can enable labs to use the massive quantities of data already accumulated and, ultimately, may be worth the up-front effort.
Automation makes headway
Automation in the form of AI or ML is entering many life science labs in a variety of ways. Researchers at Virginia Tech, for example, are designing a ML algorithm to predict the mechanics of living cells. The ability to predict shape-shifting objects like cells, which change in relation to their environment, has been particularly challenging. Their work in physics-guided machine learning aims to systematically integrate the mechanics of cell motion as biological rules and physics-based model outputs to predict the movement of cells or other shape-shifting objects in dynamic physical environments.
“Lab managers also have to gauge whether the cost and time savings are enough to justify adding AI to the lab.”
Computational scientists at Carnegie Mellon University developed a ML algorithm to understand the intricacies of genome folding in the cell nucleus and how that affects gene expression. Analyzing microscope slides may be more mundane, but can save labs significant work. “Once you have the stained slides, digitization enables AI to perform the analysis,” Fasciano says. The benefit is high throughput and greater accuracy.
A similar concept is applied to reviewing filmed experiments. Researchers at Case Western Reserve University are using ML to track the movement of flies’ wings to determine how the positioning of their halteres—hard mechanosensory organs that flap out of synchronization with the wings—affect flight. The machine learning algorithm measures the positions of the wings, the halteres, and their angles during flight by learning when their perspectives change. This application frees researchers from watching and manually recording differences from tens of thousands of frames of flies in flight.
Automation options are increasing beyond liquid handling to the processing and analysis of massive quantities of data, and AI and ML are increasingly helping scientists extract value that otherwise couldn’t be achieved quickly or, often, at all. Importantly, these trends aren’t just for large, complex labs. Instruments and applications are available now that enable even small- and medium-sized labs to benefit.