IBM RXN for Chemistry Trained with SOS Datasets
Now Available: IBM RXN for Chemistry Trained with Science of Synthesis Reaction Datasets
AI models for retrosynthesis and forward reaction prediction trained with SOS reaction datasets show significantly improved performance and higher accuracy compared to models trained on patent data alone. The IBM RXN for Chemistry platform now offers AI models trained on Science of Synthesis datasets.
RXN for Chemistry Trained with SOS Datsets Deliver Better Results
As outlined in an IBM blog post, integrating Thieme’s Science of Synthesis data has a significant positive impact on the accuracy of predictions by RXN for Chemistry for both retrosynthesis and forward reaction prediction. Thieme data is characterized by a high degree of usability for AI thanks to high-quality chemical records. Science of Synthesis data showed a usability score of 73 %. This facilitated the AI model’s learning process and resulted in more consistent predictions.
IBM RXN for Chemistry with integrated Thieme Science of Synthesis data allows users to make more accurate predictions related to new reactions in the training data set. This opens up new opportunities to explore different use cases where insights into chemical reaction patterns could enable the development of new or more sustainable products. In general, working with the SOS-trained AI tool can save time and resources since scientists can explore a wide range of reaction pathways digitally, without the need to perform physical experiments which necessitate chemicals lifecycle management from procurement to disposal. The technology thus offers environmental benefits and can make chemical reaction analyses more sustainable.
Thieme SOS Datasets
Science of Synthesis (SOS) is a reference work for organic chemistry. It contains critical reviews of the entire field of organic and organometallic chemistry. Thieme has converted this information into machine-readable format, and this highly structured data provides an important basis for training AI-based models, such as IBM RXN for Chemistry.
IBM RXN for Chemistry
IBM RXN for Chemistry is based on an AI model for neural machine translation, Molecular Transformer. This AI model derives predictions from chemical characteristics of reactants, reagents and products. When RXN for Chemistry was launched in 2018, it was trained on over 3 million chemical reactions and achieved greater than 90 % accuracy on forward predictions.
Would you like to receive a non-binding individual quotation?
If you are interested in learning more about IBM RXN for Chemistry trained with Thieme Science of Synthesis datasets, please do not hesitate to contact us. We will be happy to provide you with an offer tailored to your requirements!
Please contact: sos-datasets@thieme.com