Null reactions

TJ Tianfan Jin
QZ Qiyuan Zhao
AS Andrew B. Schofield
BS Brett M. Savoie
request Request a Protocol
ask Ask a question
Favorite

To test the model's deductive capability, a set of “null reactions” was generated that share the same reactants and reagents as real reactions but with products and input spectra corresponding to one of the reactants. Predicting the product of such reactions corresponds to identifying starting material as an unintended product using the information provided by the spectra. The introduction of null reactions also creates an underdetermined scenario for a RtP model, since a given reactant can yield multiple potential products. Null reactions were generated for each of the 299 658 real reactions. All possible null reactions were generated for reactions with multiple reactants. The USPTO dataset is large enough that some reactants are products of other reactions. In recognition of this, null reactions were discarded if their prediction target matched a real product of any reaction in the dataset. This exclusion was done to avoid accidental information leakage between null reactions and real reactions and also because it yielded a useful 2 : 1 data balance between real and null reactions without further filtering. A total of 146 672 null reactions satisfied this criteria, resulting in a combined dataset of 446 330 reactions (i.e., 146 672 null and 299 658 real) for the product identification task.

Do you have any questions about this protocol?

Post your question to gather feedback from the community. We will also invite the authors of this article to respond.

0/150

tip Tips for asking effective questions

+ Description

Write a detailed description. Include all information that will help others answer your question including experimental processes, conditions, and relevant images.

post Post a Question
0 Q&A