Publications

Physics-Inspired Machine Learning for Nanomaterial-Based Voltammetric Biosensors: A Data-Efficient Feature Engineering Approach

This study introduces a physics-inspired machine learning (ML) framework for the calibration of voltammetric nanosensing platforms. By explicitly embedding electrochemical domain knowledge, specifically dynamic resistance (R = V/I), instantaneous power (P = V × I), and logarithmic kinetic descriptors, into the ML pipeline, we enable a peak-agnostic sensing paradigm that avoids manual peak selection and baseline correction. The framework is validated across three distinct nanopolymeric biosensing systems: reduced graphene oxide–incorporated molecularly imprinted polymer (rGO-MIP) nanostructures for C-reactive protein (CRP), histidine-functionalized poly(2-hydroxyethyl methacrylate-co-ethylene glycol dimethacrylate) (HEMA-co-EGDMA) nanoparticles for vitamin C, and vitamin E acetate (VEA)-imprinted poly(2-hydroxyethyl methacrylate-co-methacryloylamido phenylalanine) (HMA-co-PA) nanoparticles for vitamin E analysis. Across 13 evaluated ML algorithms, physics-inspired features yielded a real but model-dependent improvement in predictive performance, averaging an 8.7% reduction in Test root mean square error (RMSE); the benefit was most pronounced for linear and kernel-based regressors and more limited for the strongest tree-based ensembles, which already captured substantial nonlinear structure directly from raw voltage–current signals. Permutation feature importance indicated a shift in predictive attribution toward electrochemically interpretable descriptors, though collinearity among the engineered features warrants caution in attributing this shift to any single descriptor. Grouped cross-validation and prospective external validation on independently collected replicate measurements further supported the reliability of the proposed models for new, previously uncharacterized measurements within their calibrated concentration range. A comparison with classical peak-current calibration indicated that the proposed framework performs competitively with this established approach, with its principal practical advantage lying in avoiding manual peak selection and baseline correction rather than in superior raw predictive accuracy.

Analytical applications of molecularly imprinted polymers: a personal view

Molecularly imprinted polymers (MIPs) are biomimetic counterparts of biomacromolecules that are synthesized by templating the target within a polymeric network. While the affinity of MIPs can compete with that of antibodies, the specific activity of natural enzymes remains superior to that of catalytically active MIPs. Although the mechanisms underlying the generation of antibodies and MIPs are fundamentally different, important principles have been transferred from biology to artificial binders. Templating a fragment of the target instead of the entire molecule or particle simplifies MIP synthesis and reduces costs (epitope imprinting), while hierarchical imprinting improves template orientation and generates more homogeneous binding sites than conventional random imprinting. The generation of anti-idiotypic MIPs by double imprinting follows the concept of anti-idiotypic antibodies, while post-imprinting modifications mimic post-translational modifications. Beyond classical terminal epitopes, recent approaches increasingly utilize internal epitopes, affinity tags, labels, carbohydrates, and surface motifs of viruses and cells as alternative imprinting targets. Recent advances in analytical MIP systems increasingly rely on the integration of nanomaterials to improve signal generation, mass transport, and target accessibility. In this context, nanoparticles, carbon-based nanomaterials, MXenes, and metal-organic frameworks (MOFs) have enabled the development of more efficient and functionally integrated sensing platforms. In parallel, computational rational design, artificial intelligence, and machine learning–assisted signal deconvolution are emerging as powerful tools for optimizing monomer selection, interpreting complex sensor outputs, and enabling multiplex analysis using cross-reactive MIP sensor arrays. Despite clear advantages such as higher stability, lower cost, and applicability to toxic or weakly immunogenic substances, the commercial success of MIPs is still restricted by challenges related to selectivity, reproducibility, and translation into robust real-world analytical platforms. This review discusses the evolution of analytical MIPs from classical molecular recognition materials toward increasingly intelligent biomimetic sensing systems.

Explainable AI molecularly imprinted polymer electrochemiluminescence biosensor for discriminative methamphetamine and amphetamine detection

Independent forensic quantification of methamphetamine (MA) and amphetamine (AMP) remains challenging due to their near-identical structures and overlapping electrochemical signatures. We report an end-to-end AI-integrated MIP-ECL biosensing framework employing target-specific MIP-NP/MWCNT/Nafion/Ir(pq)2(acac)/SPCE platforms. The Ir(III) luminophore within an MWCNT/Nafion matrix produced markedly superior ECL emission over the [Ru(bpy)3]2+ benchmark (40 mM DBAE, pH 8.0). MIP nanoparticles conferred high selectivity against structurally related interferents, validated in saliva, urine, and blood plasma, achieving detection limits of 0.11 ng mL−1 for MA and 0.28 ng mL−1 for AMP. A structured explainable machine learning pipeline employing CatBoost with Group-K-Fold cross-validation eliminated data leakage and achieved test R2 values of 0.9864 (MA) and 0.9675 (AMP). SHAP analysis confirmed that model predictions were consistent with known electrochemical kinetics. This modular framework establishes a generalizable paradigm for intelligent MIP-ECL biosensing with direct applicability in forensic toxicology and point-of-care diagnostics.

Unlocking the Non-Linear Frontier: A New Era of Robust Biosensing by Data Augmentation

Developing robust calibration models for electrochemical biosensors is frequently constrained by the high cost of data collection and the non-linear nature of biological signals. This small data limitation leads to overfitting in regression models and restricts the sensor's sensitivity. This study introduces an Artificial Intelligence (AI)-driven framework to enhance prediction accuracy in low-concentration ranges and stabilize sensor response without hardware modifications. For this purpose, three independent sensor datasets (C-Reactive Protein, Vitamin C, and Vitamin E) using Machine Learning (ML) algorithms benchmarked against Linear Regression were analysed. To mitigate data scarcity, four augmentation strategies were evaluated: Gaussian noise to emulate instrumental variability, statistical SMOGN, Deep Learning-based CTGAN, and interpolation based-Mixup to reinforce signal continuity. Results indicated that linear models failed (R2 < 0.10) to capture complex kinetics. While baseline ML models showed instability due to overfitting, data augmentation substantially improved generalization. Notably, physically motivated methods like Mixup and Gaussian Noise outperformed complex Deep Learning approaches, increasing Test R2 values from 0.40 to 0.65. These findings demonstrate that AI-based augmentation can effectively stabilize non-linear models and extend the quantifiable range of biosensors.

Towards a Scalable, AI-Augmented Purification Platform for Amphetamine-Type Stimulants: From Nanopolymer Synthesis to Predictive Process Simulation

The prevalent formation of by-products in conventional methamphetamine (MA) synthesis critically impairs final product purity, presenting a significant challenge that necessitates elaborate and costly downstream purification strategies. To overcome these limitations, this study introduces molecularly imprinted polymer (MIP)-based nanopolymers as a highly selective, efficient, and cost-effective alternative for the isolation of MA and its primary metabolite, amphetamine (AMP). Novel MA-imp-HEMAT and AMP-imp-HEMAT nanopolymers were synthesized via bulk polymerization and thoroughly characterized using FTIR, zeta potential, particle size analysis, and SEM. The polymers demonstrated high binding capacities of 1039 mg/g for MA and 967 mg/g for AMP with negligible cross-reactivity (relative % Q: 1.27% for MA, 2.46% for AMP). Both nanopolymers exhibited excellent reusability, retaining desorption efficiencies above 95% across five consecutive cycles. To enhance data granularity while minimizing experimental resource demands, Gaussian noise–based sampling and interpolation-driven data augmentation were applied. The fidelity of the augmented datasets was statistically confirmed through Kolmogorov–Smirnov (K-S) tests with CatBoost achieving the lowest MAE. In the AMP-non-imp-HEMAT system, SVR exhibited the best generalizability (R2 = 0.8946). This study presents a hybrid experimental–computational framework that integrates selective MIP nanopolymer design with data-driven modeling for high-fidelity prediction of analyte binding behavior. The proposed approach provides a scalable and resource-efficient alternative to conventional chromatographic purification methods, with direct implications to produce pharmaceutical-grade MA and AMP.

Beyond linear calibration: A split-and-conquer AI architecture for MIP-based electrochemical sensing of C-reactive protein

C-reactive protein (CRP) is a critical biomarker for inflammation, necessitating rapid and precise quantification for the management of cardiovascular diseases and infections. This study introduces a Physics-Informed Regional Hybrid Machine Learning-assisted electrochemical sensor by integrating previously synthesized, high-specificity CRP-imprinted poly[HEMA-co-MMA-co-(S-rGO)] nanostructures into a Nafion-modified screen-printed carbon electrode (SPCE). Although these nanostructures provided superior molecular recognition, standard linear calibration failed to resolve the complex, non-linear kinetics inherent in the voltammetric signals. To address this, we implemented a novel Split-and-Conquer AI architecture, segmenting signals into anodic and cathodic domains based on peak potential. Specialized Random Forest regressors were trained using physics-informed features (e.g., logarithmic current) on a noise-augmented dataset, achieving a robust determination coefficient of R2 > 0.94. Crucially, a High-Resolution AI Signal Enhancement Protocol (Deep Oversampling) was deployed to suppress the stochastic instrumental noise floor observed within the sensor's physical linear range (0.24–15.44 μM). This software-defined approach drastically improved sensitivity, unlocking an unprecedented Limit of Detection (LOD) of 2.5 pM and Limit of Quantification (LOQ) of 8.4 pM. The system exhibited excellent selectivity against biological interferents like BSA and uric acid, retained 80.5% signal stability after eight regeneration cycles, and achieved high ML-assisted recovery rates of 95.9–119.3% across the full dynamic range (0.543–17.39 μM) in artificial blood matrices. This study confirms that combining established MIP nanotechnology with a segmented AI framework establishes a rapid, cost-effective, and laboratory-grade platform for next-generation point-of-care diagnostics.

Polymeric Nanomaterials Containing Amino Acids for Controlled Release of Triacetyluridine

Triacetyluridine (TAU) has higher oral bioavailability and represents a novel mechanism for delivering exogenous pyrimidines to the brain. It is commonly employed to mitigate the adverse effects of chemotherapy, attenuate toxicity, enhance the effectiveness of anticancer treatment, boost cognitive functions and memory, and provide therapeutic support for neurological and hereditary disorders. p(HEMA-MAC) was synthesized using non-surfactant emulsion polymerization. Then, they were modified for TAU specificity by using immobilized metal ion (Cu+’) affinity and characterized. Optimization of TAU adsorption conditions and controlled release studies of TAU were performed. At physiological Ph = 7.4, approximately 45.8% of 2 mg/mL TAU-loaded nanopolymers were shown to be controlled, efficient release within 90 min.

Artificial intelligence-assisted design of molecularly imprinted polymers for enriching C-reactive protein

The rational design of high-affinity biomimetic materials is often hampered by the high cost of experimental research. This study presents the development of artificial intelligence-assisted molecularly imprinted polymer (AI-MIP) nanostructures for the selective recognition of C-Reactive Protein (CRP), a clinically significant biomarker associated with inflammation and infection. The development of high-purity CRP isolation systems is clinically important for accurate diagnosis, reliable disease monitoring, and effective treatment planning.

To address these challenges, CRP-imprinted nanostructures were synthesized using a surfactant-free emulsion polymerization method, which ensured clean surface properties and high molecular specificity. Methyl methacrylate (MMA) and CRP were used in the formation of a pre-polymerization complex, and subsequently, 2-hydroxyethyl methacrylate (HEMA) and silanized reduced graphene oxide (S-rGO) were incorporated to obtain the final polymeric nanomaterial. After polymerization, CRP was successfully removed from the polymer matrix, yielding the CRP-imprinted poly[HEMA-co-MMA-co-(S-rGO)] structure. CRP binding studies confirmed the selective recognition of CRP by the imprinted nanostructure. CRP-imprinted nanostructure exhibited the highest binding affinity at a CRP concentration of 1 mg/mL, reaching a maximum binding capacity, Qmax, of 350 mg/g. Characterization studies, including FTIR, SEM, and zeta size and potential analysis (317 nm,–0.213 mV, respectively), validated the successful formation, structural integrity, and colloidal stability of the synthesized nanostructure. Unlike conventional recognition elements such as antibodies or aptamers, molecularly imprinted polymers (MIPs) offer a cost-effective and chemically stable alternative for CRP isolation and purification, demonstrating selectivity and specificity toward the target molecule. To reduce the experimental burden in terms of cost, time, and resource consumption, an experimentally constrained interpolation-based data augmentation strategy was implemented, enabling the construction of high-fidelity regression models from a minimal number of laboratory trials. Among the evaluated algorithms, CatBoost Regressor achieved superior performance, accurately predicting binding behavior under varying experimental conditions. Feature importance analysis highlighted reaction time as the most influential parameter affecting CRP binding performance. This study represents, to our knowledge, the first demonstration of a CRP-selective MIP designed through integrated AI and in-silico data augmentation techniques, offering a practical and scalable pathway for affinity material discovery. The approach not only accelerates material optimization but also significantly reduces the reliance on costly trial-and-error procedures.

Development of Magnetic Nanoparticle-Added Cryogel Membrane Materials for the Removal of Pharmaceutical Wastes

The increasing use of drugs, animal husbandry, health institutions, and the pharmaceutical industry is leading to an increase in pharmaceutical residues, which can cause neurological and cardiac diseases in living organisms. Removing these wastes is crucial for ensuring living health, environment, and water safety. New-generation materials, such as cryogel membranes, can be used for drug removal through analytical separation methods combining cryogel membranes and magnetic nanoparticles. In this study, magnetic mg-Fe3O4-APTES nanoparticle-doped p(HEMA)-phenylboronic acid (PBA) cryogel membranes for the removal of dopamine (DA), which was selected as the target molecule from pharmaceutical wastes, from industrial wastewater, were synthesized, characterized, modified on the basis of boronate affinity interaction, and DA adsorption and removal studies were carried out. The developed cryogel membranes were able to adsorb DA at a rate of 81.3% in just 15 min, with a very high adsorption capacity and specificity of 1969.1 mg/g, even after five adsorption–desorption cycles, and with high specificity and specificity in an artificial urine environment prepared to simulate hospital waste. It was concluded that it performed the correct removal. In the study, specific and reusable cryogel membranes that are easy to use, low-cost and provide high-capacity removal in a short time, which can also provide physical removal with their preparation and magnetic properties, were developed. The developed cryogel membranes have the potential to be used in the field for the removal of pharmaceutical wastes with these superior properties.

Harnessing Machine Learning to Revolutionize Electrochemical Detection of Vitamin E Acetate in E-Liquids

This study presents the development of a novel molecularly imprinted electrochemical sensor for the sensitive and selective detection of vitamin E acetate (VEA) in e-cigarette liquids, a critical step in addressing the rising public health concern of e-cigarette, or vaping, product use-associated lung injury. VEA-imprinted polymeric nanoparticles, intended to serve as the recognition element on the sensor surface, were synthesized using surfactant-free emulsion polymerization. The synthesized polymer was characterized using Fourier Transformed Infrared Spectroscopy, scanning electron microscope, and zeta potential analyses. The sensor, fabricated using VEA-imprinted poly(HMA-co-PA)/Nafion on screen-printed carbon electrodes, demonstrated a limit of quantification (LoQ) of 112.3 μg/mL (3.3× S/N) with a wide linear range extending to 3.0 mg/mL (10× S/N). While the sensor exhibited limitations in detecting VEA at concentrations below the LoQ, the integration of machine learning algorithms effectively mitigated these challenges. Machine learning models successfully classified the presence of VEA, even at subdetection limit concentrations, significantly enhancing the sensor’s analytical capabilities. Rigorous testing on real-world e-cigarette liquid samples yielded high recovery rates (96.83% ± 2.79–102.56% ± 3.84), validating the sensor’s accuracy and selectivity in complex matrices. This research not only establishes a promising platform for the rapid and sensitive detection of VEA in e-cigarette liquids but also underscores the transformative potential of integrating artificial intelligence with sensor technologies for addressing critical public health challenges.

Hybrid Intelligence-Driven Nanopolymeric Sensor for Precise Electrochemical Vitamin C Analysis,
Free from LoD: Application in Real Lemon Juice

This study presents the design, synthesis, and systematic evaluation of an electrochemical nanosensor platform tailored for the precise and selective quantification of vitamin C. The sensor architecture integrates His-functionalized poly(2-hydroxyethyl methacrylate-co-ethylene glycol dimethacrylate) (His-pHEG) polymeric nanoparticles onto Nafion-modified screen-printed carbon electrodes (SPCE), thereby providing a bioactive interface with enhanced analyte affinity and stability. The His-pHEG nanoparticles were synthesized via emulsion polymerization and covalently grafted with l-histidine, as confirmed by FTIR, SEM, and zeta potential analyses. This functionalization endowed the nanoparticles with enhanced affinity and high selectivity toward vitamin C molecules, while ensuring colloidal stability and uniform morphology. Sensor fabrication parameters, including Nafion film thickness and polymer concentration, were systematically optimized to maximize electrochemical performance. The resulting His-pHEG/Nafion-modified SPCE demonstrated superior analytical characteristics, achieving a low limit of detection and a broad linear dynamic range, as determined by cyclic voltammetry and differential pulse voltammetry measurements. To overcome the fundamental limitations of conventional calibration-based electrochemical methods, such as nonlinearity and variability, a two-stage hybrid machine learning framework, specifically tailored to the inherent nature of the sensor data, was developed and integrated into the sensing workflow. The two-stage model utilized CatBoost classification to distinguish analyte presence, followed by CatBoost regression to estimate vitamin C concentration, with hyperparameter optimization ensuring robustness and predictive accuracy. Real-sample validation using lemon juice confirmed the sensor’s high recovery rates and practical applicability, demonstrating reliable performance in complex matrices. This multidisciplinary approach bridges polymer chemistry, nanotechnology, electrochemical sensing, and artificial intelligence to deliver a portable, cost-effective, and highly sensitive vitamin C detection system. Future efforts will focus on translating this platform into mobile-based, real-time analytical devices, enabling on-site applications in food quality control, healthcare, and pharmaceutical industries.

Green Carbon Dots: Synthesis, Characterization,
Properties and Biomedical Applications

Carbon dots (CDs) are a new category of crystalline, quasi-spherical fluorescence, “zero-dimensional” carbon nanomaterials with a spatial size between 1 nm to 10 nm and have gained widespread attention in recent years. Green CDs are carbon dots synthesised from renewable biomass such as agro-waste, plants or medicinal plants and other organic biomaterials. Plant-mediated synthesis of CDs is a green chemistry approach that connects nanotechnology with the green synthesis of CDs. Notably, CDs made with green technology are economical and far superior to those manufactured with physicochemical methods due to their exclusive benefits, such as being affordable, having high stability, having a simple protocol, and being safer and eco-benign. Green CDs can be synthesized by using ultrasonic strategy, chemical oxidation, carbonization, solvothermal and hydrothermal processes, and microwave irradiation using various plant-based organic resources. CDs made by green technology have diverse applications in biomedical fields such as bioimaging, biosensing and nanomedicine, which are ascribed to their unique properties, including excellent luminescence effect, strong stability and good biocompatibility. This review mainly focuses on green CDs synthesis, characterization techniques, beneficial properties of plant resource-based green CDs and their biomedical applications. This review article also looks at the research gaps and future research directions for the continuous deepening of the exploration of green CDs.