Exploring the Potential of a Multispectral-Sensing System With Automated Machine Learning for Multiplex Detection

Many analytical tests rely on colorimetric and fluorescent probes. A key factor that influences the adoption of these tests in resource-limited communities is the availability of inexpensive, user-friendly instruments able to detect their signals. In this work, we have built and tested such an instrument. It uses an off-the-shelf multispectral sensor interfaced with a Raspberry Pi Zero. We demonstrate its potential for detecting multiple analytes by monitoring the fluorescence from mixtures of quantum dots (QDs) of different colors. It was able to detect the signal from concentrations as low as 0.3 nM, making it suitable for biological applications. Instead of using complicated spectral deconvolution methods to extract the concentration of analytes present from the spectra acquired by the system, we implemented and evaluated the performance of six machine-learning (ML) algorithms for this application. We found that random forest regression (RFR) and Gaussian process regression (GPR) yielded the best fits between the predicted values and ground truths.