About

The RT-Pred predictor is an advanced machine learning tool designed to estimate retention times for small molecules across various chromatographic systems. It builds upon the Graph Neural Network Retention Time (GNN-RT), a published LC-MS RT predictor, by integrating additional machine learning algorithms. RT-Pred first categorizes molecules using void, eluted, or semi-supervised Gaussian Mixture Modeling. It then employs XGBoost to classify each molecule's category based on its structure and the chromatographic method (CM). The final prediction of retention times is performed using a graph neural network enhanced by transfer learning, which considers both the interatomic structure of the molecule and its interaction with the chromatography column. RT-Pred is highly effective, demonstrating an average coefficient of determination score (R²) of 0.95 for training datasets and 0.91 for validation datasets of CMs. The predictor accepts molecular structures in the form of the simplified molecular-input line-entry system (SMILES) and returns predicted retention times in seconds for either an existing CM or a user-defined CM system.

To create an RT prediction model for a user-defined CM, provide a description of your CM on the redirected page and upload a CSV file that includes the SMILES structures of small molecules along with their respective retention times in seconds for training. After the data is uploaded, the website will generate a trained predictor model tailored to your CM, which you can use for predicting retention times of other molecules. Additionally, you can predict retention times for molecules in existing CM methods by selecting the desired CM from the dropdown menu and uploading a CSV file with SMILES notations. This CSV file should not include a header and must consist of a single column with multiple rows, each containing the SMILES notation for one molecule. RT-Pred provides a powerful and flexible tool for accurate retention time prediction, enhancing your chromatographic analysis and research capabilities.

The server, RT-Pred, runs on a Linux system equipped with an Intel Xeon processor @ 2.20GHz and 4 CPU cores in a KVM environment. This setup is efficient for hosting backend services and ensures reliable performance for handling non-GPU-dependent tasks. It provides a stable and robust platform for deployment, ideal for applications with moderate computational requirements. The system supports web browser compatibility to ensure seamless access and functionality across various platforms. It is designed to work efficiently with modern browsers such as Google Chrome, Mozilla Firefox, Microsoft Edge, and Safari. Compatibility includes responsive design, smooth performance, and consistent rendering of the user interface across devices and screen sizes. For the best experience, using an up-to-date version of your preferred browser is recommended, as this ensures optimal security, speed, and access to the latest features supported by the application.