Machine learning guided prediction of liquid chromatography-mass spectrometry ionization efficiency for genotoxic impurities in pharmaceutical products

A Robust TabNet-Based Multi-Classification Algorithm for Infrared Spectral Data of Chinese Herbal Medicine with High-Dimensional Small Samples

Robust classification algorithms for high-dimensional, small-sample datasets are valuable in practical applications. Faced with the infrared spectroscopic dataset with 568 samples and 3448 wavelengths (features) to identify the origins of Chinese medicinal materials, this paper proposed a novel embedded multiclassification algorithm, ITabNet, derived from the framework of TabNet. Firstly, a refined data pre-processing (DP) mechanism was designed to efficiently find the best adaptive one among 50 DP methods with the help of Support Vector Machine (SVM). Following this, an innovative focal loss function was designed and joined with a cross-validation experiment strategy to mitigate the impact of sample imbalance on algorithm. Detailed investigations on ITabNet were conducted, including comparisons of ITabNet with SVM for the conditions of DP and Non-DP, GPU and CPU computer settings, as well as ITabNet against XGBT (Extreme Gradient Boosting). The numerical results demonstrate that ITabNet can significantly improve the effectiveness of prediction. The best accuracy score is 1.0000, and the best Area Under the Curve (AUC) score is 1.0000. Suggestions on how to use models effectively were given. Furthermore, ITabNet shows the potential to apply the analysis of medicinal efficacy and chemical composition of medicinal materials. The paper also provides ideas for multi-classification modeling data with small sample size and high-dimensional feature.

What is the role of current mass spectrometry in pharmaceutical analysis?

The role of mass spectrometry (MS) has become more important in most application domains in recent years. Pharmaceutical analysis is specific due to its stringent regulation procedures, the need for good laboratory/manufacturing practices, and a large number of routine quality control analyses to be carried out. The role of MS is, therefore, very different throughout the whole drug development cycle. While it dominates within the drug discovery and development phase, in routine quality control, the role of MS is minor and indispensable only for selected applications. Moreover, its role is very different in the case of analysis of small molecule pharmaceuticals and biopharmaceuticals. Our review explains the role of current MS in the analysis of both small-molecule chemical drugs and biopharmaceuticals. Important features of MS-based technologies being implemented, method requirements, and related challenges are discussed. The differences in analytical procedures for small molecule pharmaceuticals and biopharmaceuticals are pointed out. While a single method or a small set of methods is usually sufficient for quality control in the case of small molecule pharmaceuticals and MS is often not indispensable, a large panel of methods including extensive use of MS must be used for quality control of biopharmaceuticals. Finally, expected development and future trends are outlined.

Predicting liquid chromatography-electrospray ionization/mass spectrometry signal from the structure of model compounds and experimental factors; case study of aripiprazole and its impurities

A priori estimation of analyte response is crucial for the efficient development of liquid chromatography-electrospray ionization/mass spectrometry (LC-ESI/MS) methods, but remains a demanding task given the lack of knowledge about the factors affecting the experimental outcome. In this research, we address the challenge of discovering the interactive relationship between signal response and structural properties, method parameters and solvent-related descriptors throughout an approach featuring quantitative structure-property relationship (QSPR) and design of experiments (DoE). To systematically investigate the experimental domain within which QSPR prediction should be undertaken, we varied LC and instrumental factors according to the Box-Behnken DoE scheme. Seven compounds, including aripiprazole and its impurities, were subjected to 57 different experimental conditions, resulting in 399 LC-ESI/MS data endpoints. To obtain a more standard distribution of the measured response, the peak areas were log-transformed before modeling. QSPR predictions were made using features selected by Genetic Algorithm (GA) and providing Gradient Boosted Trees (GBT) with training data. Proposed model showed satisfactory performance on test data with a RMSEP of 1.57 % and a of 96.48 %. This is the first QSPR study in LC-ESI/MS that provided a holistic overview of the analyte's response behavior across the experimental and chemical space. Since intramolecular electronic effects and molecular size were given great importance, the GA-GBT model improved the understanding of signal response generation of model compounds. It also highlighted the need to fine-tune the parameters affecting desolvation and droplet charging efficiency.

Machine Learning-Enabled NIR Spectroscopy in Assessing Powder Blend Uniformity: Clear-Up Disparities and Biases Induced by Physical Artefacts

NIR spectroscopy is a non-destructive characterization tool for the blend uniformity (BU) assessment. However, NIR spectra of powder blends often contain overlapping physical and chemical information of the samples. Deconvoluting the information related to chemical properties from that associated with the physical effects is one of the major objectives of this work. We achieve this aim in two ways. Firstly, we identified various sources of variability that might affect the BU results. Secondly, we leverage the machine learning-based sophisticated data analytics processes. To accomplish the aforementioned objectives, calibration samples of amlodipine as an active pharmaceutical ingredient (API) with the concentrations ranging between 67 and 133% w/w (dose ~ 3.6% w/w), in powder blends containing excipients, were prepared using a gravimetric approach and assessed using NIR spectroscopic analysis, followed by HPLC measurements. The bias in NIR results was investigated by employing data quality metrics (DQM) and bias-variance decomposition (BVD). To overcome the bias, the clustered regression (non-parametric and linear) was applied. We assessed the model's performance by employing the hold-out and k-fold internal cross-validation (CV). NIR-based blend homogeneity with low mean absolute error and an interval estimates of 0.674 (mean) ± 0.218 (standard deviation) w/w was established. Additionally, bootstrapping-based CV was leveraged as part of the NIR method lifecycle management that demonstrated the mean absolute error (MAE) of BU ± 3.5% w/w and BU ± 1.5% w/w for model generalizability and model transferability, respectively. A workflow integrating machine learning to NIR spectral analysis was established and implemented. Impact of various data learning approaches on NIR spectral data.