Smart Vessels-On-Chips: Convergence of Artificial Intelligence and Microfluidics

Organs-on-chips: small-footprint, miniature laboratories?

The initial development of microfluidics has been inspired by progress in tissue engineering, the miniaturization of electronics, and the emergence of novel, multi-purpose materials. That allowed for precisely manipulating liquids in networks of micrometer-scale channels, changing flow rate, diffusivity, viscosity, and thermal conductivity. Dedicated materials such as polydimethylsiloxane (PDMS) can now be used to quickly fabricate chips that display chemically inert and thermostable surfaces with high biocompatibility and low toxicity to cultured cells. PDMS is also transparent, facilitating unobstructed optical analysis and allowing for morphological, spectroscopic, or fluoroscopic observation.?

Originally just a step up from traditional monolayer cell cultures, organs-on-a-chip became uber-complex devices that almost perfectly mimic cytoarchitecture in a dynamic microenvironment, allowing for the recreation of mechanobiological stimuli and gradient transport of fluids. Unlike cell or tissue slice cultures, they allow for controlled perfusion, recapitulating vascular and interstitial flows within organs. Such recreation preserves close to natural tissue and cell layers, permeable barriers, polarity, and morphological, molecular, and genetic interactions. This makes them great media for drug screening and mechanistic studies with a minimal footprint. Multiple organ models have been developed, including kidney, liver, intestine, heart, lung, brain, and tumor. [1]?

Intelligent microfluidics – organs-on-chips get smart?

Typically, results obtained from microfluidics protocol are analyzed after the experiment is completed, and AI has been applied to analyze the results. However, AI has an even greater potential for organ-on-a-chip enrichment and automated control by coupling machine learning (ML) to a microfluidic platform equipped with multiple bio-electrochemical sensors such as optical microscopy, mass spectrometry or spectroscopy. [2] Through microscopy, a time-lapse of cell migration or morphology can be recorded and analyzed, as well as spectral representation of metabolites and biomarkers. These sensors collect data fed to AI to optimize parameters and quantify the effects of biological stimuli. Based on ML, both the chip environment and external substance delivery can be modulated to achieve desirable effects. Such self-controlling, closed-loop-feedback microfluidics have a great potential to become screening devices for personalized drug response, using patient-derived organs-on-chips.?

The convergence of these technologies is challenging. In most cases, each chip will require separate training due to its heterogeneity, and the training will require supervision. The hardware is also sensitive to fouling due to tissue degradation and accumulation of metabolites that are not cleared through the device, limiting the lifespan and adding a factor of confusion to the machine learning algorithms.[1]?

Artificial intelligence used in chip closed-loop automation and data analysis is very promising and pursued by multiple researchers, but it is not the only application that can revolutionize the field.??

Hey ChatGPT, how biologically relevant are the vessels-on-chips??

In order to mimic a human organ, functional cell layers must be present, and ideally, a permeable barrier and physical dynamics should be approximated. Thus, vascularized organs-on-chips try to recapitulate the innate human physiology and anatomy to enable modeling of its pathophysiology and function. If proven translatable, such vessels-on-a-chip can satisfy a need for more predictable vascular disease models. The advancement of organs-on-chips technology allows for the creation of the endothelial lumen with the blood flow in complex vascular geometries, including stenosis, aneurysm, and bifurcation. They can then modulate hemostasis and thrombosis under arterial flow, promising a significant diagnostic value. It has been shown that platelet recruitment, aggregation, fibrin-based clotting,? and atherothrombotic processes can be modeled on such platforms.[3] Moreover, vessels-on-the-chips were used to predict anticoagulant and antiplatelet drug responses and to investigate platelet recruitment and adhesion. Even more impressive is that the microfluidic device was also used to predict the toxicity of a drug candidate that failed in the clinic but did not display such toxic vascular effects when tested prior in primates.[4]?

Vessel-on-a-chip seems to be an excellent example of a candidate for the closed-loop AI-modulated microfluidic device, where microchannel modulation can be applied to control the flow rate precisely. A library of device configuration and its influence on fluid dynamics can be analyzed by AI and applied accordingly based on the multi-sensor analysis.[1] The fundamental question remains – how vessel-like are vessels-on-chip? Despite promising results, we must remember that they are not real organs; they lack some critical anatomy, such as connective tissue and multi-organ, systemic stimuli. Artificial intelligence can help with that.?

Vascularization on a chip, a crucial organ-level feature, lacks a standardized tool or morphological metric, and traditional evaluation (vessel length, diameter, and coverage) often does not correlate with the biological function, particularly oxygen transport. Machine learning techniques, including multiple linear regression and tree-based regression (random forest), were explored to establish models predicting function based on morphology. The analysis was based on reducing morphological variable inputs and creating two new linearly combined variables with lower multicollinearity. The random forest regression model showed relatively improved predictive potential based on three critical measures: Normalized Oxygen Flux (NOF), Normalized Vascular Potential (NVP), and Normalized Oxygen Delivery (NOD). Machine learning could then provide rapid and automated predictions of vascular network function based on morphology, offering insights into the relationship between structure and function in microvascular systems. This will allow the construction of vessels-on-chips in such a configuration that they display blood flow with characteristics closely resembling actual vessels. [5]?

?

[1] Edgar A. Galan, Haoran Zhao, Xukang Wang, Qionghai Dai, Wilhelm T.S. Huck, Shaohua Ma, Intelligent Microfluidics: The Convergence of Machine Learning and Microfluidics in Materials Science and Biomedicine, Matter, Volume 3, Issue 6, 2020?

[2] Zhang YS, Aleman J, Shin SR, et al. Multisensor-integrated organs-on-chips platform for automated and continual in situ monitoring of organoid behaviors. Proc Natl Acad Sci U S A. 2017;114(12):E2293-E2302.?

[3] Gold K, Gaharwar AK, Jain A. Emerging trends in multiscale modeling of vascular pathophysiology: Organ-on-a-chip and 3D printing. Biomaterials. 2019;196:2-17?

[4] Barrile R, van der Meer AD, Park H, et al. Organ-on-Chip Recapitulates Thrombosis Induced by an anti-CD154 Monoclonal Antibody: Translational Potential of Advanced Microengineered Systems. Clin Pharmacol Ther. 2018;104(6):1240-1248?

[5] Tronolone, J.J., Mathur, T., Chaftari, C.P. et al. Evaluation of the Morphological and Biological Functions of Vascularized Microphysiological Systems with Supervised Machine Learning. Ann Biomed Eng 51, 1723–1737 (2023)?

?