Automate Chemical Data Extraction
or PDF documents and Lewis will do the rest
The need for a practical method for automated processing of image-based chemical structures
Most scientific and patent documents dealing with chemistry describe molecular structures either with systematic names or graphical images of Lewis structures. Graphical images pose inherent problems in automated processing when working with hundreds of thousands or even millions of documents since such representations cannot be directly interpreted by a computer.
Available image-based extraction methods are based on handwritten rules. A handwritten rule-based classification of structural features in molecular images is extremely error-prone because typical drawing styles of molecular structure follow no standard.
Application of Deep Neural
Network Technology
Lewis is a novel software tool based on deep neural network technology which reliably extracts machine-readable chemical structures from image representations in a convenient web platform. Instead of relying on handwritten rules, we rely on a strong combination of different neural network architectures that yield enhanced robustness.
