RT-Transformer

Key Features:

• Hybrid Deep Learning Architecture: Integrates a graph attention network for molecular graph representation with a 1D-Transformer for sequence-based feature extraction.
• Transfer Learning Across Chromatographic Conditions: Adapts a model trained on 80,038 small molecules to diverse chromatographic methods through fine-tuning on condition-specific datasets.

Scientific Applications:

• Retention Time Prediction in Metabolomics: Improves metabolite identification accuracy across multiple liquid chromatography datasets and experimental conditions.

Methodology:

RT-Transformer is trained on large-scale small molecule retention time data, learns molecular structural features via graph attention networks and 1D-Transformer layers, and applies transfer learning to recalibrate predictions for different chromatographic systems.