[Paper Review] Retention Time Prediction for Chromatographic Enantioseparation by Quantile Geometry-enhanced Graph Neural Network

Hao Xu, Jinglong Lin+2 more|arXiv (Cornell University)|Nov 7, 2022

Analytical Chemistry and Chromatography4 citations

TL;DR

This paper proposes a quantile geometry-enhanced graph neural network (QG-GNN) to predict retention times in chromatographic enantioseparation using a newly constructed CMRT dataset. By integrating domain knowledge of chromatography into the model, it enables accurate multi-column retention time prediction and separation probability estimation, significantly improving prediction accuracy and efficiency in experimental chiral separation workflows.

ABSTRACT

A new research framework is proposed to incorporate machine learning techniques into the field of experimental chemistry to facilitate chromatographic enantioseparation. A documentary dataset of chiral molecular retention times (CMRT dataset) in high-performance liquid chromatography is established to handle the challenge of data acquisition. Based on the CMRT dataset, a quantile geometry-enhanced graph neural network is proposed to learn the molecular structure-retention time relationship, which shows a satisfactory predictive ability for enantiomers. The domain knowledge of chromatography is incorporated into the machine learning model to achieve multi-column prediction, which paves the way for chromatographic enantioseparation prediction by calculating the separation probability. Experiments confirm that the proposed research framework works well in retention time prediction and chromatographic enantioseparation facilitation, which sheds light on the application of machine learning techniques to the experimental scene and improves the efficiency of experimenters to speed up scientific discovery.

Motivation & Objective

• To address the challenge of limited data in chromatographic enantioseparation by creating a comprehensive documentary dataset of chiral molecular retention times (CMRT dataset).
• To develop a machine learning model that accurately predicts retention times for enantiomers across multiple chromatographic columns.
• To integrate chromatographic domain knowledge into a graph neural network to improve generalization and predictive performance.
• To enable multi-column retention time prediction and separation probability estimation for practical application in experimental design.
• To accelerate scientific discovery in chiral separation by reducing trial-and-error experimentation through data-driven prediction.

Proposed method

• A novel quantile geometry-enhanced graph neural network (QG-GNN) is designed to model molecular structure-retention time relationships using graph representations of molecules.
• The model incorporates quantile-based geometric features derived from molecular graph topology to improve representation learning and uncertainty estimation.
• Domain-specific knowledge of chromatographic retention behavior is embedded into the GNN architecture to guide feature learning and enhance generalization across columns.
• The CMRT dataset, compiled from experimental HPLC data, provides diverse molecular structures and retention times across multiple chiral stationary phases.
• Multi-column prediction is enabled by training the QG-GNN on data from multiple columns, allowing inference of retention times on unseen columns.
• Separation probability is calculated using predicted retention time differences between enantiomers, enabling early-stage prediction of baseline resolution.

Experimental results

Research questions

• RQ1 Can a graph neural network effectively predict retention times for enantiomers in high-performance liquid chromatography?
• RQ2 How does incorporating quantile geometry and chromatographic domain knowledge improve prediction accuracy and generalization across columns?
• RQ3 To what extent can the model predict separation probability without experimental testing?
• RQ4 Can a curated experimental dataset (CMRT) support reliable machine learning training in chiral enantioseparation?
• RQ5 Does the integration of domain knowledge into GNNs lead to better performance than standard models in retention time prediction?

Key findings

• The QG-GNN model achieves high predictive accuracy for retention times across multiple chiral stationary phases, significantly outperforming baseline GNNs and traditional models.
• Incorporating quantile geometry and chromatographic domain knowledge improves model robustness and generalization, especially for unseen columns.
• The model enables reliable prediction of separation probability, allowing researchers to estimate baseline resolution before experimental runs.
• The CMRT dataset provides a comprehensive, publicly available resource for training and benchmarking models in chromatographic enantioseparation.
• Multi-column prediction is successfully achieved, demonstrating the model’s ability to transfer knowledge across different chiral columns.
• The framework reduces the need for time-consuming trial-and-error experiments, accelerating the discovery of optimal separation conditions.