Cargando…

Machine Learning-Based Quantification of (−)-trans-Δ-Tetrahydrocannabinol from Human Saliva Samples on a Smartphone-Based Paper Microfluidic Platform

[Image: see text] (−)-trans-Δ-Tetrahydrocannabinol (THC) is a major psychoactive component in cannabis. Despite the recent trends of THC legalization for medical or recreational use in some areas, many THC-driven impairments have been verified. Therefore, convenient, sensitive, quantitative detectio...

Descripción completa

Detalles Bibliográficos
Autores principales: Liang, Yan, Zhou, Avory, Yoon, Jeong-Yeol
Formato: Online Artículo Texto
Lenguaje:English
Publicado: American Chemical Society 2022
Acceso en línea:https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9434788/
https://www.ncbi.nlm.nih.gov/pubmed/36061666
http://dx.doi.org/10.1021/acsomega.2c03099
Descripción
Sumario:[Image: see text] (−)-trans-Δ-Tetrahydrocannabinol (THC) is a major psychoactive component in cannabis. Despite the recent trends of THC legalization for medical or recreational use in some areas, many THC-driven impairments have been verified. Therefore, convenient, sensitive, quantitative detection of THC is highly needed to improve its regulation and legalization. We demonstrated a biosensor platform to detect and quantify THC with a paper microfluidic chip and a handheld smartphone-based fluorescence microscope. Microfluidic competitive immunoassay was applied with anti-THC-conjugated fluorescent nanoparticles. The smartphone-based fluorescence microscope counted the fluorescent nanoparticles in the test zone, achieving a 1 pg/mL limit of detection from human saliva samples. Specificity experiments were conducted with cannabidiol (CBD) and various mixtures of THC and CBD. No cross-reactivity to CBD was found. Machine learning techniques were also used to quantify the THC concentrations from multiple saliva samples. Multidimensional data were collected by diluting the saliva samples with saline at four different dilutions. A training database was established to estimate the THC concentration from multiple saliva samples, eliminating the sample-to-sample variations. The classification algorithms included k-nearest neighbor (k-NN), decision tree, and support vector machine (SVM), and the SVM showed the best accuracy of 88% in estimating six different THC concentrations. Additional validation experiments were conducted using independent validation sample sets, successfully identifying positive samples at 100% accuracy and quantifying the THC concentration at 80% accuracy. The platform provided a quick, low-cost, sensitive, and quantitative point-of-care saliva test for cannabis.