Artificial intelligence (AI) has the potential to revolutionize laboratory operations—and for some processes, it already has. Associate editor Holden Galusha sat down with Jim Corson, PhD, vice president, life sciences solutions at Enthought to discuss how AI and machine learning (ML) are being used in cell culture labs, what lab managers should know about this technology, and how it can be implemented effectively. Note: These responses have been edited for clarity and style.
Q: Could you define the difference between artificial intelligence and machine learning?
A: AI is the broader concept of simulating human intelligence, enabling computers to perform tasks that typically require human intelligence, such as problem-solving and decision-making. ML is a subset of AI that focuses on algorithms and statistical models that allow computers to learn from and make predictions or decisions based on data, without explicitly encoding the exact result. At its heart, machine learning is essentially statistics—really complex statistics.
Q: Based on those definitions, how are AI and ML, respectively, being used in cell culture labs today?
A: AI and ML have large applications in cell culture experiments today. ML can be readily used for automating and extending image analysis workflow, which is core to cell culture experiments. ML models can automatically identify and quantify cell morphology, count cells, classify cell types, and detect anomalies or specific features of interest. These features can then be used to generate higher-order metrics such as cell viability, growth rates, and contamination to identify patterns and predict optimal conditions for cell culture, ensuring consistent quality and reproducibility.
Q: How do you envision these capabilities advancing in the future?
A: Some see the advances in AI and ML as potentially replacing researchers in the future. I see these tools as just that: tools. As AI capabilities continue to become more sophisticated, they will continue to be more valuable to scientists. Researchers will become 10 times, or even 100 times, more effective in their work by utilizing AI to operationalize their knowledge and intuition. This will allow scientists to focus more on generating new knowledge rather than merely experimental analysis and execution.
Q: How should a lab manager approach implementing these solutions in their labs?
A: Lab managers should approach implementing AI/ML solutions that have a direct, immediate impact on their researchers. This may seem obvious, but we have seen businesses try to implement AI/ML solutions that are broad, overarching, and general and those efforts are often doomed to fail. They are very expensive, take a long time to implement, and in the end, researchers don't see how it benefits them, and thus don't use them. Targeted solutions that actually are useful for a group of researchers' immediate needs are much quicker to adopt and generate momentum towards that larger AI/ML initiative.
Q: What hurdles might a lab manager face in adopting these solutions? And how can they address these hurdles?
A: One hurdle that we consistently see is researcher "trust" in the results, analyses, and recommendations of the AI models. If researchers continue to insist on manually analyzing the data to verify the model's accuracy, the AI model is not delivering any value. In fact, researchers may be actually slowed down. Verification of the model is extremely important and needs to be done as the model is being developed and deployed, but you want to quickly allow the model to begin augmenting the researchers' workflow and that requires trust. Scientists like to understand how something works (we are creatures of curiosity). Thus, one strategy that we have seen success with is educating the scientists on how AI/ML actually works. They don't need to be able to write a ML model from scratch or understand all of the complex math behind such a model. However, a broad understanding of what AI/ML can (and can't) do, will help to dispel the impression that this is a "black box of magic" that can't be trusted.
Q: What should lab managers know about using AI/ML solutions in cell culturing?
A: AI/ML algorithms rely heavily on data, so it's crucial to have a well-curated and representative dataset for training the models. Ensure that your lab collects high-quality data, including accurate and comprehensive information about cell cultures, experimental conditions, and any relevant biological parameters. Such information collection needs to be standardized across researchers to ensure that the trained models are widely applicable.
Q: What options exist for validating an AI's assessments in cell culture processes?
A: In addition to providing reliable and consistent training data, the models need to be validated by expert researchers. For the first few experiments where the model is being used, researchers should at least spot-check the results to ensure that they are indeed producing the desired results. For instance, if ML is being used to evaluate cell culture health, an expert researcher should also analyze the culture to corroborate that the culture is healthy or needs intervention. It is also important to understand that these models are not static. Developing a system where such validation information is fed back into the model to continuously improve the model will increase the longevity and accuracy of the model.
Q: How would you respond to those who are hesitant to use this technology?
A: Give it a try. There are so many open-source libraries that enable researchers to try new things with ML. All that is needed is some knowledge, data, and compute resources to get started. Having such a flexible compute infrastructure creates a low barrier to entry into the world of ML. Thus, the risk is low and the potential reward is huge. Operationalizing a machine learning pipeline takes expertise and skill, but developing a POC that such a system can work for a given research endeavor is within the grasp of any researcher.
Q: What ethical considerations are there?
A: It is important to keep the required data security and privacy measures in place for a given cell line. For instance, if these are patient-derived cell lines, ensure that the data (e.g., images, metadata, etc.) are being stored and secured appropriately.
Q: What pitfalls should a lab manager be aware of when using this tech?
A: Don't think that AI/ML should replace your scientists. We often hear lab managers wanting to implement AI/ML solutions so that they don't have to run any more expensive experiments. That is the wrong mindset. AI/ML solutions should be used to augment your research force, enabling your researchers to do more experiments, generate more data, and make better decisions faster. Think of AI/ML as a scientists' copilot. It should enable your scientists to really focus on the science, discovering the undiscovered.
Jim holds a PhD in neuroscience from the University of Virginia. His graduate work focused on the organization of sensory nerve afferents in the rodent brainstem. Jim brings a wealth of life science and software acumen to Enthought. Before joining Enthought, Jim used in vitro and in vivo electrophysiology and optogenetics to investigate the organization of neural circuits in the rodent brainstem as a postdoctoral researcher at the University of Michigan. There, he developed open-source software and hardware to control and analyze laser-scanning photostimulation synaptic mapping experiments. He also built hardware and software to control multiple 4-axis micromanipulators for multiple, simultaneous patch clamp recordings. In his free time, Jim likes to read, ride his motorcycle in the Texas hill country, and spend time with his family.