The authors present an interesting concept within this manuscript where they compare the performance of Bayesian Optimization when using One hot encoding vs other methods for the discovery of additives to improve reaction outcomes. The work presented is of interest to the readership of the journal, however, the manuscript is very light on implementation details. I would suggest the authors make the following changes and answer the below questions:

1. In Section 3.1 the authors present the various methods that they consider and use for molecular representation. However, the implementation details are summarized much later in the supplementary section. I would encourage the authors to present the implementation details within the manuscript section 3.1 itself. While I appreciate the background presented on each representation used including the advantage and disadvantages, the section also warrants concrete implementation details.
2. I especially liked Figure 2. Could this be repeated with the other three reactions as well?
3. With the small number of sample points available, I would suggest that the authors use the bootstrapped K-S test for comparing random search and Bayesian optimization (Supplementary Section 1.4).
4. Please elaborate on Supplementary Section 1.5. How does the GPR performance affect the BO search? Are there any trends that the authors have noted? Could the authors comment on why they think the BO works well despite the GPR performing rather poorly? Are there any surrogate models that the authors had tried apart from GPR?
5. Please list version numbers of all libraries and software used.

The authors have responded to all my comments and questions satisfactorily. I can now recommend this manuscript for publication.