Reproduce “MolParser-Mobile: Ultrafast OCSR System for Large-Scale Chemical Literature Mining” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Xi Fang, Haocheng Lu, Han Lyu, Chengxiang Luo, Linfeng Zhang, Guolin Ke
DP Technology developed MolParser-Mobile, an ultrafast, lightweight end-to-end Optical Chemical Structure Recognition (OCSR) system that processes 1,520 molecules per second on a single GPU. The system achieves a 38-fold speedup over previous systems while maintaining competitive or superior recognition accuracy on diverse chemical benchmarks.
Key insights
- A holistic redesign of the OCSR pipeline, integrating specialized molecule detection, large-scale real-world data, and a compact, AutoML-optimized model, is critical to overcome throughput bottlenecks without compromising accuracy.
- The encoder-decoder architecture with cross-attention efficiently separates SMILES syntax modeling from image spatial information retrieval, leading to more efficient processing of long E-SMILES strings.
- A multi-stage training approach, leveraging a diverse pre-training corpus combined with expert distillation and preference optimization, effectively mitigates pseudo-label noise and refines the recognition of complex chemical structures like Markush groups and chiral centers.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.