Pushing Differential Sensing Further: The Next Steps in Design and Analysis of Bio‐Inspired Cross‐Reactive Arrays

Abstract

Differential sensing is a technique that uses an array of crossreactive receptors to create a unique fingerprint to detect analytes. Over the past two decades significant progress in the field has highlighted the power of this approach, enabling detection with commercially available or synthetically simple sensors, discrimination of structurally similar and challenging analytes, and low concentration detection. In this Concept paper, we briefly review developments in the field and highlight areas for future exploration. We believe there is still much room to grow, particularly in designing biosensor arrays, achieving simultaneous quantification of multiple analytes (i. e., multiplexing), and implementation of more advanced machine learning algorithms in array response analysis.

Food for Thought: Optical Sensor Arrays and Machine Learning for the Food and Beverage Industry

“…36 The optical response of multiple luminescent sensor elements can be highly multiplexed in a single location, further reducing the number of array elements that need to be measured and the volume of sample required (potentially down to a whole array in one few-μL well of a 384 well plate). 25,37 The spectroscopic methods required enable the use of portable spectroscopy via simple illumination sources and basic lenses and gratings for spectroscopic analysis with a CCD or PMT photodetector, or even cellphone cameras 1,38 The analysis of the optode signals generated by the array can be achieved with the chemometric tools described above, with either point color (RGB)/wavelength changes analyzed with supervised or unsupervised discriminant or clustering methods or a full spectral analysis with partial least-squares methods. The "depth" of data versus the number of different samples and groupings is worth considering when choosing an analysis method, to ensure the method used is suitable for the data acquired (and assumptions of the method are not violated, or overfitting does not occur).…”

“…Clustering analyses (e.g., unsupervised hierarchical clustering analysis, HCA) have also been widely applied and have the advantage of defining many levels of structure or similarity in the data beyond simple nearest neighbor analysis. Regression is also increasingly valued, and methods such as partial least-squares (or projection to latent structures) regression (PLS) and orthogonal projections to latent structures (OPLS) are increasingly applied to spectral data outputs. , With the rise of increased computer power, and larger, more diverse data sets, there is also a growing move to more capable but perhaps less transparent and more easily overfitted, supervised machine learning methods including support vector machines (SVM), random forests and artificial neural nets (ANNs) …”

“…These arrays are constructed from different, optically active materials, that respond to the molecules in a sample with an optical wavelength shift (change in color), or change in luminescent intensity (either steady-state or lifetime changes) . Chromophores can be modulated via change of local environment (e.g., displacement from a supramolecular host by an analyte molecule), , or chemical reactivity between analytes and the array. , Examples include ligation of metalloporphyrins to create a change in absorbance/reflectance; the reaction or borylated fluorophores with sugars to alter their luminescence; the change of local polarity and charge around environmentally responsive fluorophores such as coumarins, fluorescent polymers or fluorescent proteins to change their color and intensity; or the change of the size, shape or local refractive index around a plasmonic nanoparticle to change their plasmonic color. − The sensor elements can be immobilized on a solid paper or glass substrate or suspended in solution before addition of the sample (typically in gas or liquid form) to generate the optical response in each array element.…”

“…The array format lends itself well to high throughput read-out methods, such as wide-field and/or hyperspectral imaging, or well-plate-style, serialized readout formats . The optical response of multiple luminescent sensor elements can be highly multiplexed in a single location, further reducing the number of array elements that need to be measured and the volume of sample required (potentially down to a whole array in one few-μL well of a 384 well plate). , The spectroscopic methods required enable the use of portable spectroscopy via simple illumination sources and basic lenses and gratings for spectroscopic analysis with a CCD or PMT photodetector, or even cellphone cameras , …”

“…23,24 With the rise of increased computer power, and larger, more diverse data sets, there is also a growing move to more capable but perhaps less transparent and more easily overfitted, supervised machine learning methods including support vector machines (SVM), random forests and artificial neural nets (ANNs). 25 Arguably the earliest examples and certainly the highest TRL (technology readiness level) sensing arrays applied in the food and beverage industry are cross-reactive electrochemical gas sensors as "chemical noses". 26 These sensors are largely based around adsorption of volatile species to arrays of chemiresistive materials, with various chemical reactivities (acidic, basic, oxygen rich/poor) and filter layers added on top.…”