A Machine Learning-Assisted Liquid Crystal Droplet Array Platform for the Sensitive and Selective Detection of Per-...

Area of Science:

• Environmental Science
• Analytical Chemistry
• Materials Science

Background:

• Per- and polyfluoroalkyl substances (PFAS) are persistent environmental contaminants.
• Accurate detection of PFAS at low concentrations is crucial for water quality monitoring.
• Existing methods for PFAS detection can be complex and costly.

Purpose of the Study:

• To develop a machine learning-assisted liquid crystal droplet array platform for sensitive PFAS detection.
• To demonstrate the platform's ability to detect specific PFAS compounds (PFOA, PFOS) at environmentally relevant concentrations.
• To explore the potential for practical environmental monitoring and differentiation of PFAS mixtures.

Main Methods:

• Utilized an autoencoder network to process images from microscale liquid crystal droplet arrays.
• Applied machine learning to analyze the latent space generated by the autoencoder for PFAS identification.
• Employed transfer learning for differentiating between PFOA, PFOS, and their mixtures.

Main Results:

• Achieved sensitive and selective detection of perfluorooctanoic acid (PFOA) and perfluorooctanesulfonic acid (PFOS) at parts-per-trillion (ppt) levels.
• Successfully detected PFAS in various water matrices, including ultrapure, tap, and simulated river water.
• Demonstrated accurate PFAS identification below U.S. EPA maximum contaminant levels, even with interfering substances present.

Conclusions:

• The ML-assisted LC droplet array platform provides a sensitive and selective method for PFAS detection in water.
• The platform shows promise for automated, high-throughput, and low-cost environmental monitoring of PFAS.
• This approach could be extended to detect other amphiphilic contaminants in real-world water samples.