Optimized De Novo Molecular Generation (OMG) for Mass Spectra Annotation Using Transfer and Reinforcement Learning.

Area of Science:

• Computational chemistry
• Artificial intelligence in chemistry
• Spectroscopy

Background:

• Assigning molecular structures to tandem mass spectra is challenging due to incomplete databases.
• Generative AI offers a solution for de novo molecular structure generation.

Purpose of the Study:

• To develop and evaluate Optimized Molecular Generation (OMG), a novel method for de novo molecular generation for mass spectra annotation.
• To improve the accuracy and efficiency of identifying chemical structures from mass spectrometry data.

Main Methods:

• Finetuning a pretrained molecular generator (REINVENT4) using transfer learning on PubChem data.
• Employing reinforcement learning with custom scoring functions for guided de novo molecular candidate generation.
• Ranking generated candidates using JESTR and ESP models.

Main Results:

• OMG achieved top-1 accuracy of 10.51% on the CANOPUS dataset and 2.42% on the MassSpecGym dataset.
• The method outperformed existing baseline approaches in spectral annotation tasks.
• Demonstrated the effectiveness of transfer and reinforcement learning for de novo generation.

Conclusions:

• Optimized Molecular Generation (OMG) significantly enhances the de novo generation of molecular candidates for mass spectra annotation.
• The integration of transfer and reinforcement learning provides a powerful framework for addressing limitations in spectral databases.
• This approach holds promise for advancing chemical structure elucidation in mass spectrometry.