Discrimination of Substandard and Falsified Formulations from Genuine Pharmaceuticals Using NIR Spectra and Machine Learning

Abstract

Near-infrared (NIR) spectroscopy is a promising technique for field identification of substandard and falsified drugs because it is portable, rapid, nondestructive, and can differentiate many formulated pharmaceutical products. Portable NIR spectrometers rely heavily on chemometric analyses based on libraries of NIR spectra from authentic pharmaceutical samples. However, it is difficult to build comprehensive product libraries in many low- and middle-income countries due to the large numbers of manufacturers who supply these markets, frequent unreported changes in materials sourcing and product formulation by the manufacturers, and general lack of cooperation in providing authentic samples. In this work, we show that a simple library of lab-formulated binary mixtures of an active pharmaceutical ingredient (API) with two diluents gave good performance on field screening tasks, such as discriminating substandard and falsified formulations of the API. Six data analysis models, including principal component analysis and support-vector machine classification and regression methods and convolutional neural networks, were trained on binary mixtures of acetaminophen with either lactose or ascorbic acid. While the models all performed strongly in cross-validation (on formulations similar to their training set), they individually showed poor robustness for formulations outside the training set. However, a predictive algorithm based on the six models, trained only on binary samples, accurately predicts whether the correct amount of acetaminophen is present in ternary mixtures, genuine acetaminophen formulations, adulterated acetaminophen formulations, and falsified formulations containing substitute APIs. This data analytics approach may extend the utility of NIR spectrometers for analysis of pharmaceuticals in low-resource settings.

Infrared Spectral Analysis for Prediction of Functional Groups Based on Feature-Aggregated Deep Learning

“…Using weight sharing, CNNs reduce the number of parameters and can extract data features of various dimensions, which can be used to identify spectral patterns and features. Two types of CNNs are used in spectral analysis: one-dimensional (1D) and two-dimensional (2D). ,, In 1D CNNs, the spectrum is treated as a sequence, and data convolution is performed based on the wavenumber dimension. ,, 2D CNNs usually need to convert spectral sequences to 2D maps , or directly take spectral images as inputs to capture correlation across wavenumbers. Based on existing methods, we want to build a model to convert spectral sequences to more informative maps for spatial feature extraction and thus take advantage of 2D CNNs more effectively.…”

“…31,33,34 In 1D CNNs, the spectrum is treated as a sequence, and data convolution is performed based on the wavenumber dimension. 32,35,36 2D CNNs usually need to convert spectral sequences to 2D maps 37,38 or directly take spectral images as inputs 39 to capture correlation across wavenumbers. Based on existing methods, we want to build a model to convert spectral sequences to more informative maps for spatial feature extraction and thus take advantage of 2D CNNs more effectively.…”

“…A DL model can automatically extract features from the data, eliminating the need for human intervention in traditional machine learning methods and allowing the algorithm to be more general. A variety of DL models have been applied to spectral analysis, including multilayer perceptron (MLP), , recurrent neural network (RNN), and convolutional neural network (CNN). ,− There are, however, limitations of each method. The MLP is composed of interleaved layers that are fully interconnected and can be overfitted .…”

Interpretable Perturbator for Variable Selection in near-Infrared Spectral Analysis

“…Multivariate calibration methods have been extensively studied for near-infrared (NIR) spectral analysis. − However, a large number of variables, often ranging in the hundreds or thousands, present in an NIR spectrum and the uninformative variables may adversely affect the model, leading to inaccurate predictions. , Variable selection that eliminates uninformative variables is necessary to simplify calibration modeling and enhance prediction accuracy and robustness. , Due to the large number of combinations of the variables and the enormous searching space, variable selection is generally a time-consuming and high-cost task. Developing effective methods for variable selection is crucial for quantitative analysis in the field of NIR spectroscopy.…”

Generalizable Descriptors of Highly Sensitive Detection of As(III) over Transition-Metal Single Atoms: A Combined Density Function Theory and Gradient Boosting Regression Approach

“…, have all contributed to the distinguished improvement of photoelectrochemical sensing, , photothermal-pyroelectric biosensor, and so on. However, although very few in situ characterizations have extensively investigated some dynamic information of chemical changes in electrochemical detection, the capture of rapid changes in the structural environment during redox reactions of sensitive materials has always been a world-class challenge and makes it urgent to refine the guiding strategies from scientific calculations. − Notably, Lieberman achieved the discrimination of substandard and falsified formulations from genuine pharmaceuticals by near-infrared spectra and machine learning (ML), while Tang’s group developed flexible and high-throughput photothermal biosensors for rapid screening of acute myocardial infarction using thermochromic paper-based image analysis . Moreover, the fiber-based sensing device of lab-on-eyeglasses has been addressed to monitor kidneys and strengthen vulnerable populations combined with the ML and density functional theory (DFT) in predicting serum creatinine using tear creatinine .…”