Markush Structures

Markush structures are generic chemical representations used in patents to claim families of related compounds. They present unique challenges for OCSR: variable substituent groups (R-groups), combinatorial enumeration, and mixed text-diagram layouts that standard molecule recognizers are not designed to handle. This group covers detection (identifying which images in a patent contain Markush diagrams) and full parsing (extracting the combinatorial structure into machine-readable form).

Paper	Year	Task	Key Idea
Detecting Markush Structures in Low SNR Images	2023	Detection	Patch-based CNN classifies whether a chemical image contains Markush structures
MarkushGrapher	2025	Parsing	Multi-modal transformer jointly encodes vision, text, and layout to parse Markush structures into CXSMILES
MarkushGrapher-2	2026	Parsing	Dual-encoder architecture with dedicated ChemicalOCR for end-to-end Markush recognition, plus IP5-M benchmark

Computational Chemistry

Dual-encoder architecture diagram for MarkushGrapher-2 showing vision and VTL encoding pipelines

MarkushGrapher-2: End-to-End Markush Recognition

An 831M-parameter encoder-decoder model that jointly encodes image, OCR text, and layout information through a two-stage training strategy, achieving state-of-the-art multimodal Markush structure recognition while remaining competitive on standard molecular structure recognition.

Computational Chemistry

Architecture diagram of the MarkushGrapher dual-encoder system combining VTL and OCSR encoders for Markush structure recognition.

MarkushGrapher: Multi-modal Markush Structure Recognition

This paper introduces a multi-modal approach for extracting chemical Markush structures from patents, combining a Vision-Text-Layout encoder with a specialized chemical vision encoder. It addresses the lack of training data with a synthetic generation pipeline and introduces M2S, a new real-world benchmark.

Computational Chemistry

Patch-based classification pipeline showing overlapping green and blue grids over a chemical image with Markush indicators highlighted in red.

One Strike, You're Out: Detecting Markush Structures

Proposes a patch-based image processing pipeline using Inception V3 to filter Markush structures from chemical documents, outperforming traditional fixed-feature (ORB) methods on low-SNR images.