Developing BioNavi for Hybrid Retrosynthesis Planning.

Area of Science:

• Computational chemistry and synthetic biology
• Artificial intelligence in chemical synthesis design

Background:

• Designing efficient and sustainable synthetic pathways is critical for producing valuable chemicals like bioactive molecules.
• Existing retrosynthesis models often neglect the integration of both chemical and biological synthesis, limiting pathway design for high-value compounds.
• The development of hybrid pathways combining chemical and biological methods offers enhanced efficiency and sustainability.

Purpose of the Study:

• To introduce BioNavi, a novel deep learning-driven model for designing hybrid synthesis pathways.
• To improve the interpretability and efficiency of synthetic pathway design by integrating multitask learning and reaction templates.
• To provide a user-friendly web server for streamlined synthetic pathway exploration.

Main Methods:

• Developed BioNavi, a deep learning model incorporating multitask learning and reaction templates.
• Evaluated BioNavi's performance against existing approaches on diverse datasets.
• Utilized case studies to demonstrate BioNavi's capability in de novo pathway design.

Main Results:

• BioNavi demonstrated superior performance compared to existing methods in pathway design.
• Achieved a 75% hit rate in successfully replicating reported biosynthetic pathways.
• Showcased significant ability in designing novel hybrid synthesis pathways.

Conclusions:

• BioNavi effectively designs hybrid synthesis pathways with high accuracy and interpretability.
• The model serves as a valuable tool for navigating and designing synthetic routes for various chemicals.
• The enhanced web server facilitates user-friendly exploration of synthetic pathway design.