Pesticide Biosensors for Multiple Target Detection: Improvement Potential with Advanced Data-processing Methods
2024 Volume 12 Pages 128-146
Details
2024 Volume 12 Pages 128-146
Pesticides are essential for agriculture, but because of their residues in crops, produce and soil, they have become a major concern. Traditional pesticide detection methods, particularly gas chromatography-mass spectrometry, are expensive and impractical for small-scale farmers. Recognizing the need for robust yet accessible sensing methods, researchers have attempted to develop pesticide sensors that use various mechanisms. Biosensors have made significant progress by utilizing biochemical mechanisms to detect trace amounts of pesticides. However, their limitations to a single pesticide or specific chemical class have driven the exploration of innovative approaches. This review covers a range of biosensor receptor and transducer mechanisms and highlights the recent advancements in biosensor technologies. The focus is on integrating modern data-processing methods, specifically for the biochemical detection of pesticides. Sensor arrays such as electronic noses (e-noses) and electronic tongues (e-tongues) are explored, along with an in-depth analysis of sensor data-processing methods, incorporating machine learning and other techniques. This review explores the potential of advanced data processing methods to effectively analyze raw sensor data from biosensor arrays, ultimately enriching the robust detection of pesticides.
Pesticides are the primary inputs in modern agriculture. It protects the plants and agricultural products from various pests and diseases, including insects, rodents, fungi, weeds, birds, and worms [1]. According to data from the Food and Agriculture Organization of the United Nations data, approximately 3.5 million metric tons of pesticides were used globally in 2021, with a steady increase in previous years [2]. Chemicals used as pesticides are designed to disrupt the regular biochemical activity of a target organism, including its nervous [3, 4], endocrine [5], and reproductive systems [6] and eventually kill or repel it. Although they are designed to harm only the pests, they also have similar harmful effects on humans, friendly insects, other animals and the environment [7, 8].
The harmful effects of pesticides increase when there is no regulation or monitoring of their use. Globally, pesticide applications are increasing significantly. Pesticide residues also enter agricultural products and ultimately into our food. Trace amounts of pesticides in food may cause multiple long- and short-term health issues, including cancer [9, 10], neurological [11] and reproductive issues [12, 13]. Hence, it is necessary to monitor the amount of pesticide residues in agricultural products to maintain food safety standards. This has converged to a need for testing pesticide residues in agricultural products.
The standard method of testing for pesticides is high-performance liquid chromatography (HPLC) [14] or gas chromatography (GC) [15] to isolate trace amounts of pesticides and mass spectroscopy (MS) [15, 16] to identify the isolated pesticide. Although these methods are accurate and can detect ultra-low concentrations of pesticides, they are expensive, time-consuming, and require specialized equipment and highly skilled operators. Hence, marginal farmers cannot test their produce to establish safety standards. Therefore, a robust, simple, inexpensive, and portable pesticide detection system is required. In addition, they should detect a wide range of pesticide classes up to the required detection limit.
Recently, biosensors have shown significant success in the detection of trace amounts of pesticides [17, 18, 19]. Biomaterials, such as enzymes, antibodies and aptamers, are used to detect and quantify the presence of pesticides. These biomaterials exhibit highly sensitive and reproducible interactions with the pesticide molecules. These sensors were designed and optimized for the detection of specific pesticides. The data generated by such sensors shows a complex relationship with various pesticides, thus they need to be explored.
Recent developments in sensor arrays and advanced machine learning (ML) methods have opened new avenues for exploring data generated by existing biosensors and discovering hidden relationships [19, 20, 21]. In this review, we explore the different directions of research on biosensors with a particular focus on data processing. Different biosensor mechanisms and data-processing methods are reviewed. Finally, we investigated the possibility of creating a robust pesticide detection biosensor array by combining the computational power of modern data-processing methods with the detection potential of novel biomaterials. This technological convergence may result in a robust sensor capable of detecting trace amounts of pesticide residues in various classes of pesticides.
A biosensor system comprises a combination of multiple functional subsystems [22]. The main subsystems are the target analytes (chemical/biological samples), receptors, transducers, data processing and visualization. They are designed to carry out the individual task in synchronization. The synchronization and symbiosis of all the subsystems are crucial for detecting the target chemical or biological material.
Figure 1 shows a block diagram of a general-purpose biosensor system. The target analyte is the substance that the biosensor is designed to detect, which can range from biological entities such as proteins and DNA to various chemical compounds such as pesticides, toxins, and pollutants. The receptor is the active site of the biosensor that interacts with the target analyte and its effectiveness depends on its ability to selectively bind to the target analyte and initiate a measurable response. This response may have various origins, including enzyme reactions or inhibition, antibodies, aptamers, direct chemical reactions, nanomaterials and biological cells. The transducer converts the measurable responses generated by the receptor into a common type of signal, generally in numeric or digital form. It mediates the interaction between the receptor and the processing system, transforming physiochemical quantities into electrical signals. The data-processing module extracts information from the raw data collected by the transducer and is responsible for classifying or estimating the concentration of the original target in the test solution based on the raw data. This may involve fixed rule-based data processing, ML and statistical methods. The data visualization module presents the processed sensor data in a visually captivating and comprehensible format, enabling users to interpret and analyze the information effortlessly.
These components work together to enable biosensors to detect and quantify target analytes accurately, making them valuable tools in various fields, including healthcare, environmental monitoring and food safety.
Pesticide biosensors are specialized devices designed to detect and quantify pesticides in various samples. They utilize biological components to interact with target pesticides and produce measurable signals to determine pesticide concentrations. The development of sensitive and selective detection methods is crucial because of the wide variety of pesticide types and their ultra-low concentrations. Researchers have improved pesticide biosensors in three main ways: detection mechanisms at the receptor level, compatible transducers and advanced data processing. The detection mechanism mimics the binding properties of pesticides with specific receptors, such as enzymes [23], antibodies [24] and aptamers [25, 26], for precise and reliable detection at ultra-low concentrations. Compatible optical [27, 28, 29] and electrochemical [30, 31, 32] transducers convert biochemical activity into detectable standard signals with high accuracy and repeatability. Data-processing techniques, including ML, are employed to interpret and extract meaningful information from collected sensor data, enabling the accurate analysis and prediction of the behavior of a target analyte. Pesticide biosensors offer efficient analysis, greater precision, detection at lower concentrations, continuous monitoring and costs lower than traditional pesticide detection methods, making them valuable for pesticide detection in various applications [18, 33, 34, 35].
2.3 Biomaterials for pesticide detection: enzymes, antibodies and aptamers| Biomaterial (Enzyme/ Antibody/ Aptamer) | Target | Transduction | Range | Limit of Detection (LoD) | Ref |
|---|---|---|---|---|---|
| AChE | OP pesticides, specifically malathion | Electrochemical | 0.01–1 ng/mL | 2.6 pg/mL | [36] |
| Carbofuran | Electrochemical | [23] | |||
| Methyl parathion | Electrochemical | 1–2 ppm | 0.48 ppb | [31] | |
| Chlorpyrifos and carbaryl | Electrochemical | 0.5 ng/mL in real samples. | [38] | ||
| OP pesticides | Electrochemical | 0.5–100 ng/mL (1.73–345.7 nM) | 0.18 ng/mL (0.62 nM) | [37] | |
| AChE and choline oxidase (CHO) | OP pesticides: parathion-methyl | Raman scattering | 5×10−9–5×10−4 M | 1.7×10−9 M | [39] |
| Organophosphorus hydrolase (OPH) | Paraoxon | pH changes | 0.5 μg/mL | [40] | |
| Ethyl paraoxon and methyl parathion |
Colorimetric and fluorometric detection |
0.014 mM Ethyl paraoxon, 0.044 mM methyl parathion | [41] | ||
| Alkaline phosphatase (ALP) | Methyl paraoxon | Fluorescence and electrochemical | ≈ 0.65 nM | [42] | |
| OP pesticides, specifically chlorpyrifos | Fluorescence | 20 pg/mL–1000 ng/mL | 15.03 pg/mL (S/N = 3) | [29] | |
| 2,4-dichloro-phenoxy-acetic-acid | Electrochemical | 0.04–24 nM | 16 pM | [43] | |
| Glyphosate antibody | Glyphosate | Electrochemical | 10 ng/mL–50 ug/mL | 10 ng/mL | [32] |
| Glyphosate antibody and atrazine antibody | Glyphosate and Atrazine | Electrochemical | 0.5 ng/mL–10 μg/mL for glyphosate and 10 fg/mL–1 ng/mL for atrazine | 0.5 ng/mL for glyphosate and 1 fg/mL for atrazine | [44] |
| Anti-OPs-McAb antibody | OP: Parathion-methyl, fenitrothion, and fenthion | Optical: Localized surface plasmon resonances (LSPR) | 104 to 5 × 106 pg/mL for parathion-methyl and fenitrothion, 5 × 104 – 107 pg/mL for fenthion | [45] | |
| Monoclonal antibody (Anti-ChlT-mAb) | Chlortoluron | Optical | 0.01–10 µg/L | 22.4 ng/L | [24] |
| Glyphosate antibody | Glyphosate in human urine | Electrochemical | 0.1–72 ng/mL | 0.1 ng/mL | [30] |
| Tetramethylrhodamine labeled aptamer | OP: Phorate, profenofos, isocarbophos and omethoate | Fluorescence | 0.333, 0.167, 0.267 and 0.333 µg/L | [28] | |
| FenA2 DNA aptamer | Fenitrothion | Fluorescence | 14 nM (3.88 ppb) | [26] | |
| SS2-55 aptamer | Isocarbophos | Lateral flow | 2.48 μg/L | [25] | |
| DIAZ-02 ssDNA aptamer | Diazinon | Fluorescence | [46] | ||
| DNA aptamer | Malathion | Optical: Liquid crystal | 0.465 nM | [47] |
Recent advancements in biosensor technology have led to a paradigm shift in pesticide detection, with a notable focus on enzyme- and antibody-based biosensors. Biosensors exploit the specificity of enzymes and antibodies to detect and quantify pesticide residues in environmental, agricultural and food samples. The integration of these biorecognition elements with innovative transduction methods has enhanced the sensitivity, selectivity, and efficiency of biosensors, thereby offering promising solutions for accurate and rapid pesticide monitoring.
Many enzymes, antibodies and aptamers have been used to develop biosensors targeting distinct pesticide groups. Table 1 lists the recently developed sensors with comparative sensor parameters. Enzymes, such as acetylcholinesterase (AChE) [36, 37, 38] and organophosphorus hydrolase (OPH) [40, 41] have been used to detect organophosphate (OP) pesticides through inhibition and catalytic hydrolysis reactions, respectively. Antibodies, particularly monoclonal antibodies [24], exhibit high affinity and specificity, making them ideal for recognizing specific pesticide residues. Their interactions with various pesticide groups, including organochlorines, pyrethroids, and triazines, forms the basis of highly selective biosensing platforms. Recently, aptamers have emerged as potential receptors for highly selective biosensors. These DNA- or RNA-based biosimilar materials can be designed and synthesized easily and sustainably, opening a new avenue for custom receptors to detect pesticides [25, 26] or similar targets.
Multiplexing is a process of combining multiple sensing elements to detect one or more targets. In biosensors, arrays of sensing elements with different enzymes, antibodies, or other active materials are combined into a single sensor (also known as a sensor array). The primary objective of multiplexing in biosensors is to achieve the simultaneous detection of multiple targets; however, this method can also improve the sensing quality [48]. Researchers have achieved improved sensitivity [49, 50], selectivity, repeatability, detection range, and compatibility with complex sample matrices by using sensor multiplexing through sensor arrays. The limitations of a single sensing element are compensated by other sensing elements in the same array, according to the desired sensor design [22].
Simultaneous detection of multiple targets in biosensors is achieved by utilizing an array of sensing elements, where an individual element is designed to detect various targets. This approach enhances the capability of the sensor to identify and quantify various substances concurrently, thereby providing a broader range of applications in medical diagnostics, environmental monitoring and food safety. The simultaneous detection of multiple targets using a sensor array is illustrated in Fig.2. This explains the flow of the detection process from the test sample to the final results. Multiplexing improves the overall quality of detection in the following areas:
・Selectivity is the ability of a biosensor to distinguish between a target molecule and other molecules (with related chemical, physical, or biological properties). In a sensor array, the simultaneous detection of multiple targets prevents cross- selectivity and improves the overall selectivity of the sensor [52]. A higher selectivity facilitates the precise detection of analogous molecules.
・Reproducibility is another essential parameter that defines the ability of a sensor to produce the same output when subjected to the same input. Processing redundant sensor data from multiple sensors, with various fault-correction [53] algorithms remove possible errors in sensor readings and can provide reproducible results. This improves the reliability of the sensor.
・The range of detection refers to the range of target analyte concentrations that can be quantified by the sensor. The data from multiple sensing elements with various detection ranges are merged to obtain a more comprehensive detection range. A biosensor requires a more comprehensive detection range that covers the minimum and maximum limits according to the requirements.
・The compatibility with complex sample matrices is crucial for sensor operation in real-world scenarios. The real samples contained a wide range of chemicals (sample matrices), solution pH values [54] and temperatures. Single sensors must be highly standardized and optimized. Alternatively, the effects of the sample matrices, pH, and temperature in the sensor array are negligible by processing redundant and sufficient data from individual sensing elements using an appropriate data processing algorithm. This is a significant boost to the sensor performance and reliability in real-world applications.
Sensors (or sensor arrays) generate raw numerical data. This represents various parameters, including the absorbance, RGB intensity, current and potential, depending on the type of sensor (optical or electrochemical) used. The information hidden in the numerical data also varies with sensing type and sensor design.
Optical sensors use absorbance, fluorescence, or RGB intensity to determine the biochemical reaction in the test solution to measure the target marker’s concentration representing the target concentration. Numerical changes in absorbance or fluorescence provide a precise and quantitative readout. In this method, the detected molecule does not decompose during detection, which makes re-examination of the same experiment possible, such as the online measurement of swimming pool water chlorine concentration [55]. Recent developments in portable photometric systems have made them suitable for field deployment [56]. A simplified version of the optical system uses a paper or an object similar to the base. Target biomolecules are detected by measuring color changes [57] visually or using color sensors/cameras to quantify color changes resulting from specific biochemical reactions [58, 59, 60]. Paper was used as the common substrate for these sensors [61, 62]. Data from these paper-based sensors can be recorded by a machine (camera or color sensors) and analyzed by visual comparison with a standard color gradient [63]. This makes it simple and field deployable. The intensities of the red, green and blue colors of the sensing element are recorded as the sensor data for further analysis.
Finally, electrochemical sensors produce numeric electrical signals during reactions, ensuring precise biomolecular detection through distinct signatures. This method measures the electrochemical potential or current while keeping the other parameters constant. Based on the experimental design, various scanning methods exist, such as cyclic voltammetry [64], differential pulse voltammetry [65] and chrono-amperometry. Electrochemical methods can detect ongoing oxidation-reduction reactions where the concentration of a target is measured by the oxidation or reduction peak of the target molecule.
4.2 Requirement of advanced data processing in sensor data analysisA regular sensor provides more straightforward data with linear or logarithmic relationships between the raw data and final result. The relationship between the raw data and results is well known, stable and reliable. Therefore, a fixed rule can be implemented to obtain results from the raw data, as shown in Fig. 3(A). For example, in the Bradford method [66], raw data (absorbance at 595 nm) are directly proportional to the analyte concentration. Absorbance measures the binding between a dye (Coomassie brilliant blue G-250) and protein sample. The protein concentration is calculated by comparing the test absorbance to a standard curve generated by the reaction of known amounts of a standard protein, typically bovine serum albumin.
However, the collected raw data are sometimes complex, and the actual relationship between the raw data and results is unknown. Consequently, the fixed rule-based system cannot generate results from such raw data. As shown in Fig. 3(B), advanced ML techniques help extract features from complex multivariable data from the sensor array. An example is an E-nose sensor for fruit-ripening [10]. Unrelated raw data (analog voltages from multiple sensors) are analyzed to classify the different stages of fruit ripening.
4.3 Utility of sensor data processingData processing is an essential step in obtaining insights from the data collected from the sensor array. Agricultural food sensors can detect a wide range of parameters using diverse detection mechanisms. Hence, sensor data are also complex. Data processing is performed in multiple directions, including calibration, error handling, classification, regression, prediction, compensation and feature minimization.
Calibration is the process of correlating raw sensor data with the actual target parameters. Error handling detects and manages the situation when the raw data from the sensor shows an error or are out of the normal range. Classification algorithms are then used to classify the input sensor data into different fixed classes of parameters. Similarly, regression algorithms are used when a secondary parameter is calculated using data from multiple sensors as the primary input. Prediction is another important use of data processing, in which future parameters are predicted based on current sensor data. One typical example of such a system is “yield prediction of crops” [67]. In addition, prediction systems are used to estimate the missing data from a sensor. Finally, feature minimization reduces the large amount of data in the sensor array to a more compact form by removing redundant or uncorrelated data. This is important for reducing sensor bias and drift and simplifying the model. Diverse algorithms and data processing mechanisms are used to achieve these utilities.
4.4 Algorithms for sensor data processingFor the processes mentioned above, various algorithms have been implemented in sensors used in agriculture. Standard algorithms for agricultural food sensors include artificial neural networks (ANN), fuzzy logic, linear discriminant analysis (LDA), indicator displacement assay (IDA), principal component analysis (PCA), k-nearest neighbor (k-NN) and support vector machines (SVM). The use of these methods in bio sensors is summarized in Table 2. These algorithms are dynamic and can handle multiple roles based on the system design.
An ANN is a powerful tool with excellent predictive power. It imitates the function of biological neurons to perform calculations. These networks excel in tasks, such as predicting crop yields [87, 88] and diagnosing diseases [89]. Their ability to process complex relationships within data sets makes them highly valuable in decision-making processes and contributes significantly to advancements in the field. Because of their complexity, they have evolved into more advanced versions such as deep neural networks (DNN) [90], convoluted neural networks (CNN) [91], and recursive neural networks (RNN) [92]. ANN-based processing is used to detect dangerous gases, such as methane, carbon monoxide [71], SO2, NO2, H2, ethanol [72] and blood glucose levels from breath analysis [73]. Advanced versions, such as DNN and CNN, are used to detect vitamin B6 derivatives [74] and RNN is widely used in medical diagnoses, such as processing electrocardiogram (ECG) electrode data to detect arrhythmia [75]. Drone image-based agricultural yield prediction has also been successfully performed using a CNN [67].
| Use of data processing | Sensors | Target | Ref. |
|---|---|---|---|
| ANN and least squares regression (LSR) | SnO2 gas sensors | Dangerous and odorless gases, such as CH4 and CO | [71] |
| ANN | SnO2 gas-sensor array | Propane-2-ol, methanol, acetone, ethyl methyl ketone, hexane, benzene, and xylene | [68] |
| ANN | GaN sensor array | SO2, NO2, H2, ethanol and their mixtures | [72] |
| ANN | Quartz crystal microbalance sensors | Blood glucose level from breath analysis | [73] |
| DNN, CNN | 3D fluorescence spectra | Vitamin B6 derivatives | [74] |
| RNN | ECG data | Arrhythmia disease | [75] |
| Fuzzy Logic | MOS sensor | Indoor air quality | [76, 77] |
| CNN, BpNN | Multi-metal sensor array | Rice quality | [69] |
| LDA, SVM, and partial least squares regression model | pH, free acidity, hydroxymethylfurfural content, and electrical conductivity | Honey adulteration | [78] |
| IDA | Five colorimetric metal-indicator complexes | Antibiotics in food | [27] |
| PCA | NPK, pH, and conductivity | Determination of the N, P, and K in soil | [79] |
| k-NN | Colorimetric sensor array | Black tea fermentation | [80] |
| PCA and k-NN | Air pollution in outdoor environments, namely benzene, toluene, xylene (BTX), NO2, and CO | Thin-film sensor array | [81] |
| PCA-SVM algorithm | Functionalized PEDOT: PSS | K+, Ca2+, and Mg2+ ions | [82] |
| SVM | Gas sensor array | Varroosis mite | [83] |
| SVM | Gas sensor array | Onion sour skin | [84] |
| SVM | MOS sensor | Formaldehyde in mixed VOCs | [85] |
| Multilayer perceptron, random forest, and extreme gradient boosting models | Seven gold electrodes coated with layer-by-layer films of poly (o-methoxy aniline), poly (3-thiophene acetic acid), and molybdenum disulfide (MoS2) | Bisphenol A, estrone, and 17-β-estradiol and their mixtures | [86] |
| Decision tree models | Chitosan, chondroitin sulfate, sericin, and gold nanoparticles/sericin-based sensing unit | Staphylococcus aureus and bovine mastitis in milk | [70] |
Fuzzy logic [93] can efficiently manage imprecise information in control systems under various environmental conditions. The unique ability of fuzzy logic to navigate uncertainties and imprecision enhances decision-making in diverse aspects of agriculture and food sciences. Fuzzy logic contributes to more accurate and reliable outcomes by providing a nuanced approach to data interpretation, particularly in decision-making. Owing to its simple but efficient decision-making ability, it is used in indoor air quality sensors [76, 77].
LDA [94], IDA and PCA [95] are helpful in revealing patterns within complex data. These algorithms are necessary for classification and feature minimization tasks to ensure accurate and insightful results. They help to isolate meaningful information from large datasets with many inputs and are typically used to detect honey adulteration [78], antibiotics in food [78] and N-P-K in soil [79].
The k-NN [96, 97] algorithms contribute to proximity-based decision-making. Its capability to classify data points based on their similarity to neighboring points has been harnessed in various applications, offering efficient and accurate categorization in sensor data analysis. As an ML algorithm, it is beneficial for training data for classification. It is used to detect various targets such as the fermentation level of black tea [80] and air pollutants [81].
The SVM [98] is a robust tool for classification tasks in agriculture. Their ability to define optimal decision boundaries ensures the accurate categorization of sensor data, particularly in scenarios where precision is paramount for effective decision-making. The SVM with PCA detects K, Ca and Mg ions [82] and formaldehyde in mixed VOCs [85]. It is also directly used to detect sour skin in [84] and Varroa mites [83].
In recent years, more complex relationships have been discovered between the raw sensor data and target parameters. Improvements in the aforementioned algorithms have been widely acknowledged. Improving the processing capability also aids in implementing the abovementioned algorithms in small and portable electronic devices, allowing them to be deployed in the field.
4.5 Data processing in pesticide and toxicity biosensorsData processing in a pesticide sensor array becomes more complex [99], because a low target concentration makes the raw signal small and the number of chemical groups is also high. Therefore, more attention is required when processing pesticide sensor data. Robust calibration and classification are required to achieve better responses [100]. Although this is a recent trend, researchers have used various data-processing tools to extract insights from raw sensor data to detect pesticides or other toxic chemicals. Table 3 lists examples of pesticide or toxicity sensors that use data-processing-based detection methods. Data-processing methods such as PCA, LDA, SVM, ANN, k-NN, and multivariate statistical methods were used in various combinations to detect pesticide residues such as curathane, numetrin and nativo (in water) [102], thion [103] and diazion (in sweet cherries) [104].
| Target | Sensor | Data processing technology used | Ref |
|---|---|---|---|
| Curathane, numetrin and nativo in water | Carbon screen-printed electrodes | PCA, LDA, SVM, k-NN, Naïve Bayes | [102] |
| Thion pesticides such as malathion, parathion, chlorpyrifos and diazinon | Six metal nanoparticles were spotted on a piece of filter paper | Visual detection, multivariate statistical methods | [103] |
| Odorous air and headspace | MOS gas sensors, metal electrodes | PCA, hierarchical cluster analysis, SVM | [101] |
| Diazinon residue in sweet cherries | MOS Sensors: TGS813, RGS822, TGS2602, MQ3 | ANN, PCA, LDA | [104] |
| Cypermethrin and chlorpyrifos | 10 MOS sensors in a PEN3 electronic nose | PCA, LDA, SVM | [105] |
A The combination of sensor arrays and advanced data processing in multi-target sensors is an emerging trend. This could lead to a sophisticated and robust system for detecting a wide range of chemicals, including pesticides. Figure 4 shows the interactions between different targets and multiple assays to produce a final result using a data-processing model. The array comprised multiple sensors, each interacting with a sample to generate raw data from the individual sensors. The system effectively analyzes complex patterns to obtain a final result using advanced data-processing techniques such as PCA and SVM. This analytical approach enables precise discrimination between various pesticide compounds in the sample. The outcome is the creation of a comprehensive pesticide composition profile, offering valuable insights for informed decision-making in agriculture and environmental monitoring. Through this systematic and advanced approach, this sensor system may significantly enhance the efficacy and accuracy of pesticide detection, thereby facilitating optimal agricultural and environmental management.
Tang et al. [105] also developed a pesticide detection system with an E-nose sensor array to detect pesticide residues on apples using 10 metal oxide semiconductor (MOS) gas sensors and processed the data using PCA, LDA, and SVM.
Voss et al. [106] developed a MOS gas sensor array-based system to detect the ripening status of peaches. Individual sensors in a sensor array cannot directly detect volatile gases released during ripening. However, they successfully determined the ripening status by processing raw sensor data using the random forest method with LDA.
5.2 Data processing to optimize sensor responseThe optimization of the sensor response through data processing can result from modern technological advancements, significantly contributing to improved accuracy, reliability and efficiency across various applications. This process involves the application of advanced algorithms and techniques designed to analyze and refine raw data acquired by sensors, calibration adjustments, dynamic model adjustments, and data fusion. The goal is to extract meaningful information from input signals while mitigating the impact of external factors that may introduce noise or inaccuracies.
ML and advanced data analysis techniques refine sensor data and extract sensitive and dynamic information hidden in the raw data, which has future potential to replace the existing rule-based data processing methods. These techniques enhance the accuracy and alignment with measured characteristics, making them crucial in medical devices and scientific instruments. For example, the use of the causal Kalman filtering method and neural networks for blood glucose monitoring [107] has improved the accuracy and reliability of diabetes management.
ML techniques are increasingly used for calibration adjustments in biosensor arrays to improve sensor accuracy and reliability. Random forest-based algorithms have enhanced low-cost sensor calibration strategies, particularly for weather sensors in urban areas [107]. Temperature sensors [108] can also be calibrated using ML algorithms to address feature differences. Hg2+ sensors [109] also exhibited improved performance at low concentrations. These examples demonstrate the potential of ML in ensuring sensor accuracy and reliability in industrial settings.
The dynamic model adjustment automatically modifies the response based on the input. ML algorithms transform sensor responses, which is a significant benefit. Sensors use neural networks to learn from previous data and adapt to changing conditions. This adaptive capability improves the ability of the sensor to detect complex patterns and anomalies, resulting in a more accurate response. For example, the dynamic model of amperometric biosensors allows for comprehensive signal output characterization with errors of < 3% and no significant deviations from the glucose detection output curve [110]. Calibration of dynamic models is essential for ML-based hypertension risk assessment. Accurate hypertension risk assessment requires calibration, particularly when combining photoplethysmography and ECG data [111]. These examples demonstrate the importance of dynamic model adjustment and calibration to ensure the accuracy and reliability of biosensor data for signal characterization and health risk assessment.
Data fusion, particularly in biosensor data processing, combines multiple data sources to improve understanding and accuracy. Biosensor data insights depend on advanced data processing algorithms such as ML. For example, ML algorithms can recognize complex biological process patterns (e.g., food intake episodes) in biosensor signals from wearable sensors [112]. ML techniques can analyze biosensor data streams in real time to detect biological sample anomalies using data fusion [77]. They can predict and optimize biosensor performance by learning from historical data and improving biosensor-based measurements. These examples demonstrate how data fusion and advanced data processing methods such as ML can transform biosensor technology for better analytical results.
5.3 Challenges and strategy for field-deployable pesticide detection biosensorA comprehensive strategy is required to develop a reliable, field-deployable pesticide detection biosensor. This ensures the efficacy, versatility and usability in real-world scenarios. Critical considerations to ensure precise and reliable results include continuous monitoring, cost-effectiveness, in-field operability, rapid screening of food contaminants, a broad detection range and a low limit of detection (LoD). Combining a sensor array with advanced data processing is an excellent approach to achieving this goal.
Continuous monitoring and in-field operability are critical components of this approach. Portable electrochemical transduction devices allow for the continuous monitoring and rapid assessment of pesticide levels without the need for elaborate laboratory setups. The integration of the Internet of Things (IoT) enables real-time data transmission, remote monitoring and management of biosensors across large agricultural landscapes. Another critical aspect of this strategy is the cost-effectiveness of solutions based on electrochemical transduction. During in-field deployment, electrochemical transduction provides a low-cost solution optimized for energy efficiency using integrated embedded systems.
Biosensors with electrochemical transduction and IoT are used for the rapid screening of food contaminants. This combination enables quick and accurate screening processes, allowing for quick decisions regarding pesticide contamination. Enzymes and substrates are carefully selected using electrochemical principles to ensure target analyte specificity and multi-pesticide detection. ML and ANN contribute significantly to the determination of specific pesticides, allowing for versatile adaptability. Electrochemical biosensors combined with ML and ANN can be tailored to accommodate different food matrices. This integrated approach offers a comprehensive solution for agricultural applications, ensuring biosensor efficacy and adaptability to various agricultural ecosystems.
In response to growing concerns regarding pesticide residues, this review emphasizes the importance of innovative and accessible detection methods. However, traditional techniques, such as HPLC and GC-MS are unsuitable for small-scale farmers. Biosensors containing biomaterials such as enzymes and antibodies meet the requirement for portable, low-cost, and versatile pesticide detection systems capable of identifying a wide range of pesticide classes. Our research covers a wide range of biosensor mechanisms, including electrochemical and optical approaches and sensor arrays such as E-noses and E-tongues. Data-processing methods, including ML techniques, was the focus of this review. Combining biosensor array technology with modern data processing methods is a promising path for developing powerful pesticide-detection biosensor arrays. Biosensors generate complex but valuable information, opening new possibilities, and recent advances in sensor arrays and ML provide unprecedented opportunities to decipher hidden relationships within a complex data landscape. This review proposes a significant shift in pesticide detection strategies with the aim of developing an advanced biosensor array capable of detecting trace amounts of various pesticide groups. This development will help make agricultural landscapes safer and more sustainable.
According to Richardson et al. [113], six out of nine boundaries for a livable Earth have already been exceeded. Therefore, it is crucial to take action to improve planetary health. Monitoring tools, such as portable pesticide sensors, are one way to achieve sustainability in agriculture. This review may be a small step towards sustainable agriculture and a healthy Earth.
We thank the Japan Educational Exchanges and Services (JEES) for the T-Banaji scholarship, which allowed KC to study in Japan as a PhD student. We are grateful to Prof. Naoto Ogawa and Prof. Atsuhiro Shimada, and all members of the Biomolecular Chemistry Laboratory for their valuable suggestions and continuous support. We are also thankful to Editage (www.editage.jp) for the English language editing.
