Integrating machine learning for rapid and accurate multiplex identification of the allelic variants in single...

Single nucleotide polymorphisms (SNPs) are widely used in precision medicine, disease predisposition assessment, nutrigenetics and authenticity testing of agricultural and food products. SNP genotyping is much more challenging than detecting longer DNA sequences because it must distinguish allelic variants differing by just one base pair. Lateral flow genotyping assays offer portability, simplicity, low cost, speed and preloaded reagents. A growing trend is toward multiplex assays capable of identifying multiple allelic variants simultaneously. However, interpreting multiple test zones is time-consuming, subjective, and prone to errors. Thus, multiplexing often compromises both practicality and accuracy. To address these challenges, we integrated machine learning for rapid and accurate multiplex genotyping. The proposed workflow includes DNA isolation, multiplex PCR, a 15-min multiplex genotyping reaction, multiallele lateral flow analysis, and machine learning-based signal interpretation for full genotype acquisition. As a model system, we studied the identification of eight allelic variants across four SNPs in Olea europaea, using Convolutional Neural Network (CNN) and Decision Tree algorithms. The critical phase of algorithm training and testing was conducted using constructed genotypes made from oligonucleotide mixtures, eliminating the need for multiple DNA isolations and PCR steps, thereby simplifying and streamlining the process. CNN was selected for its superior performance, achieving 100 % genotyping accuracy for all constructed genotypes, while the Decision Tree model ranged from 67 % to 100 %. In real DNA samples, genotyping accuracy ranged from 93 % to 100 %. Overall, the proposed method achieved 97 % accuracy in identifying the cultivar origin of DNA from olive oil and olive leaves.