Science of Synthesis on IBM RXN for Chemistry

Science of Synthesis Reaction Datasets

Enter the organic chemistry matrix: high-quality reaction data, encoded for machines

Choose Your Reality!

The SOS datasets can help you in both real and virtual laboratories

Ready-to-use model or simply the datasets

Choose between a pre-trained model or the datasets for training your algorithm

Best Methods, Best Results

A joint study with IBM showed that supplementing a large, publicly available reaction dataset extracted from patents with the SOS data when training AI models for retrosynthesis and forward-reaction prediction leads to significant improvements in performance.

Models Trained with SOS Datsets Deliver Better Results

Analysis of the results by an international group of leading synthetic chemists showed that when compared to the model trained on the patent data alone, the IBM RXN model trained on a combination of the patent data with the SOS dataset gave improved accuracy in forward-reaction prediction by a factor of 3, and improved performance in retrosynthesis by a factor of 9. The models trained on the Science of Synthesis data are now available to you on the IBM RXN for Chemistry platform.

Contact Our SOS Dataset Experts

If you are interested in further information, please get in touch. We will be pleased to provide you with our content! Send an email to sos-datasets@thieme.com