Revolutionizing microfluidics with artificial intelligence: a new dawn for lab-on-a-chip technologies

Keisuke Goda

Keisuke Goda presently serves as a professor in the Department of Chemistry at the University of Tokyo. In addition, he holds adjunct professorships in the Department of Bioengineering at UCLA and the Institute of Technological Sciences at Wuhan University. His academic credentials include a summa cum laude B.A. in physics from UC Berkeley and a Ph.D. in physics from MIT. Goda's research team is on a mission to pioneer ‘serendipity-enabling technologies’. The goal is not only to expand the frontiers of human knowledge and understanding, but also to mentor global leaders who are expected to define the future trajectory of biology and medicine.

Hang Lu

Hang Lu is the C. J. “Pete” Silas Chair and Professor of Chemical and Biomolecular Engineering at Georgia Tech. Her current research interests are microfluidics, machine learning and quantitative analyses, and their applications in neurobiology, cell biology, cancer and biotechnology. Her awards and honors include an ACS Analytical Chemistry Young Innovator Award, NSF CAREER award, Alfred P. Sloan Foundation Research Fellowship, DARPA Young Faculty Award, and MIT Technology Review TR35 innovator. She is an elected fellow of the AAAS, AIMBE, and RSC (UK). Her lab's work has been and is supported by the NSF, NIH, private foundations and others.

Peng Fei

Dr. Peng Fei is a professor at Huazhong University of Science and Technology. He is a physicist who develops advanced microscopes for life science. He has published over 70 peer-reviewed papers in leading journals, such as Nature Methods, Nature Materials, Nature Communications, PNAS, Optica, eLife, etc., and held over 40 granted patents and software copyrights. These research outcomes have been highlighted by several professional scientific media, such as News & Views at Nature Photonics and Nature Methods. Dr. Fei has given over 40 invited talks at several renowned academic conferences, such as SPIE Photonics West and OSA Biophotonics Congress, and wrote several book chapters for Tissue Optics.

Jochen Guck

Jochen Guck received his PhD in Physics from UT Austin in 2001. After academic positions at the University of Leipzig, Cambridge University, and Technische Universität Dresden, he is now Director at the Max Planck Institute for the Science of Light, in Erlangen, Germany. His research centers on exploring the physical properties of biological cells and tissues, their link to cell function, and potential clinical applications. He has authored over 200 peer-reviewed publications and 8 patents. His work has been recognized by several awards, including the Cozzarelli Award by the National Academy of Sciences, the Paterson Medal by the Institute of Physics, and an Alexander-von-Humboldt Professorship.

In our swiftly changing scientific and technological landscape, the marriage of two distinct yet complementary fields, microfluidics and artificial intelligence (AI), is garnering considerable attention (https://doi.org/10.1039/D2LC00813K).^1–5 This groundbreaking blend of physical manipulation and digital intelligence signifies a crucial turning point for lab-on-a-chip technologies, marking the onset of an innovative epoch of scientific exploration and technological advancement. This unique synergy opens up opportunities for notable advancements in a myriad of sectors, ranging from medical diagnostics and personalized therapeutics to environmental monitoring and complex biochemical research. In today's world, where the quest for efficiency and miniaturization is relentless, the integration of AI and microfluidics is set to revolutionize our approach to these challenges (https://doi.org/10.1039/D2LC00813K).^1–5 It provides solutions that transcend traditional innovation, offering transformative potential that could alter the landscape of many scientific and technological fields.

Microfluidics, the discipline focused on manipulating and controlling fluids on the micro- and nano-meter scale, has already brought about a significant revolution in biological and chemical research methodologies.^6–8 The miniaturization achieved by microfluidics has dramatically reduced the need for large sample and reagent volumes, leading to cost reductions and accelerated reaction times. These advantages have solidified the position of microfluidics as a fundamental component of lab-on-a-chip technologies. It enables high-throughput, parallelized analysis, opening up new opportunities in point-of-care diagnostics, drug discovery and environmental testing, among other applications. Nevertheless, the journey of microfluidics from its conceptual inception to widespread application has encountered a series of challenges.^9,10 Complexities associated with intricate device fabrication, precise fluid control, comprehensive data analysis, and overall system optimization require sophisticated and robust solutions.

This is where the promise of AI comes into play. Armed with capabilities such as pattern recognition, predictive analytics, and autonomous decision-making, AI offers an arsenal of tools that can effectively navigate the hurdles faced by microfluidics while enhancing its capabilities (https://doi.org/10.1039/D2LC00813K).^1–5 AI, particularly machine-learning algorithms, can optimize the design of microfluidic devices, augmenting their efficiency, adaptability, and ability to perform complex tasks (https://doi.org/10.1039/D2LC00254J, https://doi.org/10.1039/D2LC00322H).^3,11–14 Ranging from relatively simple tasks such as flow control and mixing, to more complex functions such as sorting and detection, AI can fine-tune these processes, enhancing their reliability and accuracy. Furthermore, AI's inherent ability to learn from data and adapt to new situations can aid in the development of “smart” microfluidic devices (https://doi.org/10.1039/D2LC00254J, https://doi.org/10.1039/D2LC00843B). These devices can respond to changing conditions in real time, a feature that could prove extremely useful in dynamic environments such as in vivo diagnostic or therapeutic applications.

In the realm of microfluidics, one of the most significant applications of AI is data analysis (https://doi.org/10.1039/D2LC00813K).^1,4,15–17 High-throughput screening in microfluidic experiments often generates an overwhelming deluge of data. However, AI, with its superior processing power and pattern recognition capabilities, can effectively navigate this sea of information, identify patterns and correlations, and provide valuable insights to guide decision-making processes (https://doi.org/10.1039/D1LC01006A, https://doi.org/10.1039/D2LC00984F, https://doi.org/10.1039/D1LC00755F).^18–28 In contrast to conventional data analysis methods, which may overlook subtle patterns within these extensive datasets, AI has the potential to illuminate hidden connections, thus enabling researchers to gain fresh insights and drive unprecedented discoveries (https://doi.org/10.1039/D2LC00983H, https://doi.org/10.1039/D1LC01043C, https://doi.org/10.1039/D2LC00596D, https://doi.org/10.1039/D2LC00902A).^29–36 Beyond mere analysis, the capacity of AI to extrapolate these insights for predictive modeling carries profound implications for microfluidic applications. AI can construct models based on past data to forecast experimental outcomes or system behaviors, thereby empowering researchers to anticipate results and optimize their experiments proactively.† This predictive capability proves particularly crucial in arenas such as drug discovery and personalized medicine, where accurately predicting a drug's efficacy or a patient's response to treatment can save precious time, cut costs, and in some cases, even save lives.

The fusion of AI and microfluidics represents a revolutionary change in point-of-care diagnostics, indicating a significant potential to redefine healthcare as we currently understand it.‡^37,38 AI-powered microfluidic systems have the ability to swiftly analyze biological samples such as blood and saliva, delivering accurate diagnoses in real time.§ This advancement is particularly pivotal in resource-limited settings and during infectious disease outbreaks, where rapid diagnosis and prompt treatment are paramount (https://doi.org/10.1039/D1LC00467K, https://doi.org/10.1039/D2LC00478J). Yet, the revolution does not stop at diagnostics. AI-integrated microfluidic devices also have tremendous potential in the realm of personalized medicine. They can harness patient-specific data to customize treatments, with AI algorithms analyzing data from these devices in real time, thus enabling healthcare providers to monitor and adapt a patient's treatment regimen as necessary (https://doi.org/10.1039/D2LC00304J, https://doi.org/10.1039/D2LC00637E). This capability is particularly beneficial in managing chronic illnesses, such as diabetes or cardiovascular disease, where continuous monitoring and treatment adjustments can significantly enhance patient outcomes. All in all, the fusion of AI and microfluidics, encapsulating everything from diagnostics to personalized treatment, could democratize healthcare, making advanced diagnostics and personalized treatments accessible to all, regardless of geographical or economic constraints.

Indeed, while AI assists microfluidics, the latter also contributes significantly to AI (https://doi.org/10.1039/D2LC00813K).^1–5 Microfluidics can bolster AI in a number of ways. Firstly, microfluidic devices excel at producing vast quantities of high-quality data. They have the ability to execute a multitude of experiments or tests concurrently, maintaining precise control over conditions. Such data can subsequently be employed to train AI algorithms, thereby enhancing their prediction accuracy and decision-making abilities. Secondly, microfluidic systems commonly come equipped with sensors that are adept at monitoring conditions in real time. This sensor-derived data can be introduced into an AI system, which allows it to adapt the system's operations based on fluctuating conditions. Such dynamic adaptability can dramatically boost the efficiency and versatility of AI systems. Thirdly, microfluidics is known to facilitate automation of complex procedures and unifies multiple experimental steps into one device. When coupled with AI, this has the potential to spawn fully automated systems that can undertake intricate tasks with minimal to zero human intervention. Lastly, due to their small size and comparatively low production cost, microfluidic systems are very appealing. They offer an avenue to apply AI in scenarios where it might be considered unfeasible due to constraints related to size or cost.

In sum, the dynamic integration of AI and microfluidics heralds a new dawn for lab-on-a-chip technologies, marking a transformative breakthrough that magnifies the potential of each individual discipline. This convergence is more than just the sum of its parts – it is a synergistic interaction that promises to overcome the existing challenges in the field, enhancing efficiency, and driving the boundaries of what is currently achievable with lab-on-a-chip technologies. As we continue to delve into and comprehend the interplay between AI and microfluidics, we are on the precipice of a scientific revolution that extends beyond research into healthcare and other sectors. This union paves the way for the development of smarter, more adaptive devices that can learn from data, adapt to changing conditions, and make intelligent decisions. We are entering a new era where diagnostics, therapeutics, and research are not just personalized but also predictive and proactive. The future of lab-on-a-chip technologies is indeed bright, powered by the transformative confluence of AI and microfluidics. The thematic collection “AI in Microfluidics” showcases some of the most promising first steps into this future.

Acknowledgements

This paper was written with the assistance of ChatGPT, which was instrumental in revising the initial draft.

Notes and references

1. A. Isozaki, et al. , Lab Chip, 2020, 20, 3074–3090.
2. L. Liu, et al. , Nanoscale, 2021, 13, 19352–19366.
3. E. A. Galan, et al. , Matter, 2020, 3, 1893–1922.
4. J. Riordon, et al. , Trends Biotechnol., 2019, 37, 310–324.
5. J. Zheng, et al. , Biosens. Bioelectron., 2021, 194, 113666.
6. G. M. Whitesides, Nature, 2006, 442, 368–373.
7. E. K. Sackmann, et al. , Nature, 2014, 507, 181–189.
8. S.-Y. Teh, et al. , Lab Chip, 2008, 8, 198–220.
9. S. Battat, et al. , Lab Chip, 2022, 22, 530–536.
10. D. T. Chiu, et al. , Chem, 2017, 2, 201–223.
11. N. H. Bhuiyan, et al. , Anal. Chem., 2022, 94, 3872–3880.
12. N. Pouyanfar, et al. , Trends Environ. Anal. Chem., 2022, 34, e00160.
13. S. Srikanth, et al. , Sens. Actuators, A, 2021, 332, 113096.
14. Y. Ding, et al. , Anal. Chem., 2020, 92, 132–149.
15. S. Z. Harofte, et al. , Small, 2022, 18, 2203169.
16. B. Wang, et al. , Nat. Commun., 2023, 14, 1341.
17. K. C. M. Lee, et al. , Trends Biotechnol., 2021, 39, 1249–1262.
18. N. Nitta, et al. , Cell, 2018, 175, 266–276.
19. A. A. Nawaz, et al. , Nat. Methods, 2020, 17, 595–599.
20. H. Zhang, et al. , Biomed. Opt. Express, 2019, 10, 1044–1063.
21. Y. Zhou, et al. , Trends Biotechnol., 2021, 39, 978–989.
22. A. Isozaki, et al. , Lab Chip, 2020, 20, 2263–2273.
23. H. Mikami, et al. , Nat. Commun., 2020, 11, 1162.
24. K. Chung, et al. , Nat. Methods, 2008, 5, 637–643.
25. M. M. Crane, et al. , Lab Chip, 2010, 10, 1509–1517.
26. M. M. Crane, et al. , Nat. Methods, 2012, 9, 977–980.
27. M. M. Crane, et al. , Lab Chip, 2009, 9, 38–40.
28. M. Zhan, et al. , PLoS Comput. Biol., 2015, 11, e1004194.
29. S. Chaudhary, et al. , Nat. Commun., 2022, 13, 5165.
30. T. Zhu, et al. , Sens. Actuators, B, 2021, 348, 130638.
31. Z. Wang, et al. , Nat. Methods, 2021, 18, 551–556.
32. S. Chaudhary, et al. , Nat. Commun., 2022, 13, 5165.
33. S. Chaudhary, et al. , eLife, 2021, 10, e60321.
34. A. San Miguel, et al. , Nat. Commun., 2016, 7, 12990.
35. K. Bates, et al. , PLoS Comput. Biol., 2022, 18, e1009942.
36. I. Carlos de Cáceres, et al. , Integr. Biol., 2018, 10, 539–548.
37. M. Herbig, et al. , Sci. Rep., 2022, 12, 870.
38. D. Soteriou, et al. , Nat. Biomed. Eng., 2023 DOI:10.1038/s41551-023-01015-3.

Footnotes

† https://doi.org/10.1039/D2LC00938B, https://doi.org/10.1039/D1LC01087E, https://doi.org/10.1039/D2LC00376G, https://doi.org/10.1039/D2LC00206J, https://doi.org/10.1039/D1LC00467K, https://doi.org/10.1039/D2LC00416J, https://doi.org/10.1039/D2LC00764A.

‡ https://doi.org/10.1039/D2LC00764A, https://doi.org/10.1039/D1LC00467K, https://doi.org/10.1039/D2LC00028H, https://doi.org/10.1039/D2LC00084A, https://doi.org/10.1039/D2LC00478J.

§ https://doi.org/10.1039/D1LC01140E, https://doi.org/10.1039/D2LC00166G, https://doi.org/10.1039/D2LC00482H, https://doi.org/10.1039/D2LC00289B.

---