RetroRanker: leveraging reaction changes to improve retrosynthesis prediction through re-ranking
Problem
Data-driven retrosynthesis models suffer frequency bias, producing low-ranked, chemically unreasonable reactant predictions.
Why it matters
Faulty retrosynthesis predictions can mislead chemists planning drug and chemical syntheses.
What the AI did
RetroRanker, a graph neural network re-ranking model, re-ranked predictions of existing retrosynthesis models using potential reaction changes to demote chemically unreasonable candidates.
Results
not stated in source (abstract truncated before benchmark figures)
Impact
Re-ranking approach proposed to mitigate frequency bias; gains not quantified in provided abstract.
Record
| Venue | Journal of Cheminformatics |
|---|---|
| Published | 2023 |
| Work type | article |
| Authors | Junren Li, Lei Fang, Jian–Guang Lou |
| Citations | 10 (OpenAlex, at import time) |
| References | 42 |
| Model or method | RetroRanker (graph neural network ranking model) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1186/s13321-023-00727-7 |
| Open access | gold |
| Keywords | retrosynthesis, re-ranking, graph neural networks, frequency bias, chemistry |
Importance rating
real but narrower advance
Retrosynthesis matters for organic chemistry and synthesis planning, but this is a re-ranking refinement of existing models; abstract gives no scale of impact.