Gas Chromatographic Retention Index Prediction Using Multimodal Machine Learning

Abstract

Gas chromatography is a widely used method in analytical chemistry and metabolomics. Using gas chromatography, vaporizable compounds can be separated for their further identification. Retention indices are standardized values that depend only on a chemical structure of a compound and on a stationary phase and characterize the retention of a compound in a chromatographic system. Retention index prediction is an important task because databases contain experimental values for a small fraction of all possible molecules, while this information is usable for untargeted analysis. In this work, we consider four machine learning models for retention index prediction: 1D and 2D convolutional neural networks, deep residual multilayer perceptron, and gradient boosting. String representation of the molecule, 2D representation of the chemical structure, molecular descriptors and fingerprints, and molecular descriptors are used as inputs of these four models, respectively, along with information about the stationary phase. The first and third models show the best performance, while the other two perform slightly worse. The models predict retention index values for various standard and semi-standard non-polar stationary phases. Further improvement in performance was achieved using a linear model that uses the results of four previous models as inputs (model stacking). The models were tested using various diverse data sets: flavor compounds, essential oils, metabolomics-related compounds. Achieved accuracy: median absolute and percentage errors -6-40 units and 0.8-2.2%. Accuracy depends on a test data set. The stacking model outperforms previously reported approaches for all test data sets. Parameters of a pre-trained model and some source code are provided.INDEX TERMS Analytical chemistry, convolutional neural network, deep learning, gas chromatography, gradient boosting, residual neural network, retention index, untargeted chemical analysis.

Deep Learning Based Prediction of Gas Chromatographic Retention Indices for a Wide Variety of Polar and Mid-Polar Liquid Stationary Phases

“…CNN and MLP are convolutional neural network and multilayer perceptron, respectively, trained using data for standard and semi-standard non-polar SP. The structure and hyperparameters of these neural networks are very close to the models used in work [17]. Two other deep learning models, referred to as CNNPolar and MLPPolar, are the same as CNN and MLP but are trained with RI for standard polar SP.…”

“…An important task is the development of versatile RI prediction models that are applicable to almost arbitrary structures. There are several works (e.g., the most recent works [17][18][19]) that are devoted to RI prediction for diverse compounds and use data sets ranging in size from hundreds to tens of thousands of compounds. Most of such works, except for the most recent ones, are extensively reviewed in our previous work [17].…”

“…Deep learning models are often much more accurate than models based on hard-coded features when large enough training sets are available. Over the last two years, at least four works [17][18][19]21] applied deep learning methods for RI prediction using large (tens of thousands compounds) data sets. Another work [22] uses deep learning for the prediction of GC retention time of some steroids.…”

“…For example, a popular functional group contributions-based model [23] gives a median absolute error of 46. QSRR models [24] trained using the Flavornet database (https://www.flavornet.org/, accessed on 22 August 2021) provide mean and median errors of about 50 [17] for non-polar SP. For deep learning models, mean and median errors below 30 and 20, respectively, are achievable for the same or similar data sets [17][18][19].…”

“…Almost all "versatile" RI prediction models [17] are developed for non-polar SP. There are several such works for polar SP (e.g., polyethylene glycol (PEG), DB-WAX).…”