READRetro: Natural Product Biosynthesis Planning with Retrieval-Augmented Dual-View Retrosynthesis
Lee, S.; Kim, T.; Choi, M.-S.; Kwak, Y.; Park, J.; Hwang, S. J.; Kim, S.-G.
Show abstract
Elucidating the biosynthetic pathways of natural products has been a major focus of biochemistry and pharmacy. However, predicting the whole pathways from target molecules to metabolic building blocks remains a challenge. Here we propose READRetro as a practical bio-retrosynthesis tool for planning the biosynthetic pathways of natural products. READRetro effectively resolves the tradeoff between generalizability and memorability in bio-retrosynthesis by implementing two separate modules; each module is responsible for either generalizability or memorability. Specifically, READRetro utilizes a rule-based retriever for memorability and an ensemble of two dual-representation-based deep learning models for generalizability. Through extensive experiments, READRetro was demonstrated to outperform existing models by a large margin in terms of both generalizability and memorability. READRetro was also capable of predicting the known pathways of complex plant secondary metabolites such as monoterpene indole alkaloids, demonstrating its applicability in the real-world bio-retrosynthesis planning of natural products. A website (https://readretro.net) and open-source code have been provided for READRetro, a practical tool with state-of-the-art performance for natural product biosynthesis research.
Matching journals
The top 9 journals account for 50% of the predicted probability mass.
Similar papers in this journal
- A Transferable Deep Learning Approach to Fast Screen Potent Antiviral Drugs against SARS-CoV-2 94%
- Predicting anti-cancer drug synergy using extended drug similarity profiles 93%
- Enhanced compound-protein binding affinity prediction by representing protein multimodal information via a coevolutionary strategy 91%
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.