(195d) Endotoxin Sensors Using Liquid Crystals and Machine Learning | AIChE

(195d) Endotoxin Sensors Using Liquid Crystals and Machine Learning


Jiang, S. - Presenter, University of Wisconsin-Madison
Smith, A., University of Wisconsin - Madison
Abbott, N. L., Cornell University
Noh, J., Cornell University
Zavala, V. M., University of Wisconsin-Madison
Endotoxins are lipopolysaccharides that are found in the outer membranes of gram-negative bacteria and underlie many health hazards such as endotoxemia (a septic shock caused severe immune response) [1]. A number of methods have been proposed to detect the presence of endotoxins with remarkable sensitivity [2,3,4]. However, the most commonly used methods rely on complex reagents derived from the blood of crabs, do not provide a quantitative estimation of concentration, and/or are incapable of classifying endotoxin type.

Liquid crystal (LC) droplets dispersed in water are sensitive to the presence of endotoxins and undergo internal ordering transitions when exposed to different endotoxin species and concentrations [5]. Specifically, endotoxin accumulates at topological defects within the LC droplets, triggering a change from droplet configurations with bipolar symmetry to droplet configurations with radial symmetry. Different internal configurations of LC droplets have distinct optical properties [6]. As such, flow cytometry of LC droplets can be used to generate rich forward scattering/side scattering (FSC/SSC) data. A variety of counting methods used in the analysis of FSC/SSC data can predict endotoxin concentration [7], however, the optimal method for characterizing the scatter plots is not known and some methods are subject to variability.

In this work, we demonstrate that machine learning techniques can be used to accurately predict endotoxin concentrations and species from FSCC/SSC data. Our approach uses a convolutional neural network to extract pattern information (features) from FSC/SSC density data. Our framework reveals that significant hidden information is available in FSC/SSC data. Our analysis also provides insights into the physical effects driving changes in the LC internal configurations.


[1] R. A. Proctor, Handbook of Endotoxin. Elsevier, 1984.

[2] T. Muta, T. Oda, and S. Iwanaga, “Horseshoe crab coagulation factor B. A unique serine protease zymogen activated by cleavage of an Ile-Ile bond.,” Journal of Biological Chemistry, vol. 268, no. 28, pp. 21384–21388, 1993.

[3] K. G. Ong, J. M. Leland, K. Zeng, G. Barrett, M. Zourob, and C. A. Grimes, “A rapid highly-sensitive endotoxin detection system,” Biosensors and Bioelectronics, vol. 21, no. 12, pp. 2270–2274, 2006.

[4] T. Y. Yeo et al., “Electrochemical endotoxin sensors based on TLR4/MD-2 complexes immobilized on gold electrodes,” Biosensors and Bioelectronics, vol. 28, no. 1, pp. 139–145, 2011.

[5] N. Bao, Y. Zhan, and C. Lu, “Microfluidic Electroporative Flow Cytometry for Studying Single-Cell Biomechanics,” Analytical Chemistry, vol. 80, no. 20, pp. 7714–7719, 2008.

[6] I.-H. Lin, D. S. Miller, P. J. Bertics, C. J. Murphy, J. J. de Pablo, and N. L. Abbott, “Endotoxin-Induced Structural Transformations in Liquid Crystalline Droplets,” Science, vol. 332, no. 6035, p. 1297, Jun. 2011.

[7] D. S. Miller, X. Wang, J. Buchen, O. D. Lavrentovich, and N. L. Abbott, “Analysis of the Internal Configurations of Droplets of Liquid Crystal Using Flow Cytometry,” Anal. Chem., vol. 85, no. 21, pp. 10296–10303, Nov. 2013.