These are preliminary reports that have not been peer-reviewed. They should not be regarded as conclusive, guide clinical practice/health-related behavior, or be reported in news media as established information. For more information, please see our FAQs.
Preprints are manuscripts made publicly available before they have been submitted for formal peer review and publication. They might contain new research findings or data. Preprints can be a draft or final version of an author's research but must not have been accepted for publication at the time of submission.
submitted on 15.10.2019 and posted on 21.10.2019by Frederik Sandfort, Felix Strieth-Kalthoff, Marius Kühnemund, Christian Beecks, Frank Glorius
Despite their enormous potential, machine learning methods
have only found limited application in predicting reaction outcomes, as current
models are often highly complex and, most importantly, are not transferrable to
different problem sets. Herein, we present the direct utilization of Lewis
structures in a machine learning platform for diverse applications in organic
chemistry. Therefore, an input based on multiple fingerprint features (MFF) as
a universal molecular representation was developed and used for problem sets of
increasing complexity: First, molecular properties across a diverse array of
molecules could be predicted accurately. Next, reaction outcomes such as
stereoselectivities and yields were predicted for experimental data sets that
were previously evaluated using (complex) problem-oriented descriptor models. As
a final application, a systematic high-throughput data set showed good
correlation when using the MFF model, which suggests that this approach is
general and ready for immediate adoption by chemists.