Case Study: Friedrich Schiller University Jena achieves faster, more accurate chemical structure recognition with Google Cloud Platform

Friedrich Schiller University Jena How Deep Learning is helping to understand life at the molecular level

Friedrich Schiller University Jena, a public research university in Germany, was struggling to process fragmented chemistry data and train deep learning models large enough to convert chemical structure images into machine-readable formats. The university’s Steinbeck research group needed far more compute than its existing GPUs could provide, and turned to Google Cloud Platform with Cloud TPUs to speed up its open-source chemistry tools, including DECIMER and STOUT.

Google Cloud Platform helped the team move their scripts to TPUs through the TPU Research Cloud, cutting training times from months to days and enabling training on extremely large datasets. The results were significant: model training became up to 16x faster, a 400-million-image model was trained in two months instead of more than a year, and tool accuracy improved to over 90% for segmentation and up to 96% for image transformation.

Friedrich Schiller University Jena

Christoph Steinbeck

Professor for Analytical Chemistry