Insights into predicting small molecule retention times in liquid chromatography using deep learning

Area of Science:

• Computational metabolomics
• Analytical chemistry
• Bioinformatics

Background:

• Untargeted metabolomics uses liquid chromatography-mass spectrometry (LC-MS) for small molecule identification.
• Accurate metabolite annotation is hindered by the vast number of small molecules and challenges in predicting retention time (RT).
• In silico tools exist but often yield extensive candidate lists, necessitating improved RT prediction for accurate identification.

Purpose of the Study:

• To review advancements in AI-driven retention time (RT) prediction for small molecules in computational metabolomics.
• To highlight the impact of large RT datasets, particularly the METLIN dataset, on deep learning applications in RT prediction.
• To discuss databases, molecular representation methods, AI algorithms, and challenges in AI-based RT prediction for metabolite annotation.

Main Methods:

• Review of scientific literature focusing on AI and deep learning for RT prediction in metabolomics.
• Analysis of publicly available RT datasets and molecular representation techniques.
• Discussion of AI algorithms applied in recent RT prediction studies.

Main Results:

• AI, particularly deep learning, has shown significant progress in RT prediction accuracy over the past five years.
• The availability of large RT datasets has been crucial for training effective AI models.
• AI-based RT prediction shows promise in reducing false positives and facilitating metabolite annotation.

Conclusions:

• AI technology offers powerful solutions for improving small molecule RT prediction in metabolomics.
• Further research is needed to address molecular representation inconsistencies and achieve practical, widespread application of AI tools.
• Continued development in AI algorithms and datasets will enhance the accuracy and efficiency of metabolite identification.