Questions about inconsistencies between the paper and the released data

[LittletreeZou]

Thank you for integrating and opensource the Benckmark dataset.
I noticed that there are some inconsistencies between statistics in the paper and the released data in benchmarks/CodonBERT/data. Here are the confusing parts:

• For the MLOS flu vaccine data, you show 543 mRNA samples in Table 1 in the paper, but I only found 167 samples in the released data.
• For SARS-Cov-2 vaccine degradation data, you show 2400 mRNA samples in Table 1 in the paper, but I only found 233 samples in the released data.

Could you kindly clarify them?

BTW, I noticed that some of the datasets are very small. When using a 0.7/0.15/0.15 split on such a small dataset and computing metrics like correlation, the results are not reliable. It would be better that you use k-fold cross validation.

[phil-fradkin]

To follow up on this maybe we can restrict the scope of the question to the consistency with the datasets.

In downloading the data I found:
(Downloaded - Reported)

MLOS: 167 - 543
TC Riboswitches: 355 - 355
CoV Vaccine: 2400 - 2400
mRFP Expression: 1459 - 1459
Fungal Expression: 7089 - 7056
E. Coli Proteins: 6348 - 6,348
mRNA Stability: 65,356 - 41,123

It would be helpful if the authors clarified the length discrepancies between mRNA stability, Fungal Expression, and MLOS datasets.

Thanks a lot and congrats on the publication!