Deep convolutional neural network for retention time prediction in Reversed-Phase Liquid Chromatography
Repository contains pre-trained models, data on retention times, one-hot matrices for five data sets (METLIN SMRT, MassBank1, MetaboBASE, Hilic_Retip and Riken_Retip). Scripts can be used to train 1D CNN model from scratch or for transfer learning approach. To reproduce results of predicition retention times for METLIN SMRT data set with 1D CNN check Report.
- The main script to build model with METLIN SMRT data set and use it for transfer learning on your own data set is "FINAL SCRIPT FOR TRANSFER LEARNING". Load data from ZIP file "Data for training+trasfer script.zip"
- To train 1D CNN on METLIN SMRT data set load zip file "Train initial model for SMRT"
- To train 1D CNN from scratch on LIFE_old, LIFE_new, MassBank1, MetaboBASE, Hilic_Retip, and Riken_Retip data sets load zip files "List of matrices transfer data sets" and "SMILES and RTs"
- For transfer learning with LIFE_old, LIFE_new, MassBank1, MetaboBASE, Hilic_Retip, and Riken_Retip data sets load zip files "List of matrices transfer data sets", "SMILES and RTs", "Pre-trained"
- DATA also contains SDF files forLIFE_old, LIFE_new, MassBank1, MetaboBASE, Hilic_Retip, and Riken_Retip data sets "Molecules.zip". These files can be used to build your own model and to compare results with 1D CNN
The only requirements are to be familiar with the basic syntax of the R language, PC with Internet connection and Windows OS (desirable), RStudio and R (≥ 4.0.0).
SMRT_transfer has been published in the Journal of Chromatography A. If you use this software to analyze your own data, please cite it as below, thanks:
Please send any comments or questions you may have to the author (Ms. Elizaveta Fedorova 👩🔬): ✉️ elizaveta.chemi@gmail.com, 