Recently, I tried to reproduce the results of ESM-1v in the PABP_YEAST_Fields2013-singles dataset, but I found that when performing zero-shot inference on PABP_YEAST_Fields2013-singles according to the method of esm/examples/variant-prediction/predicts, the best spearman results of five esm-1v model were only 0.51, much lower than the zero-shot result 0.6658 reported in the paper. And ESM-1b is also only 0.60, lower than 0.68 which report in the paper.
In BLAT_ECOLX_Ranganathan2015 dataset I using esm-1b can get the result of spearman 0.70 which is similar to the result 0.72 in the paper.
Could you please tell me how to reproduce the results in the paper? Do I need to do some processing on the results of the five esm1v outputs, or calculate the spearman with the label separately? Thank you.
Recently, I tried to reproduce the results of ESM-1v in the PABP_YEAST_Fields2013-singles dataset, but I found that when performing zero-shot inference on PABP_YEAST_Fields2013-singles according to the method of esm/examples/variant-prediction/predicts, the best spearman results of five esm-1v model were only 0.51, much lower than the zero-shot result 0.6658 reported in the paper. And ESM-1b is also only 0.60, lower than 0.68 which report in the paper.
In BLAT_ECOLX_Ranganathan2015 dataset I using esm-1b can get the result of spearman 0.70 which is similar to the result 0.72 in the paper.
Could you please tell me how to reproduce the results in the paper? Do I need to do some processing on the results of the five esm1v outputs, or calculate the spearman with the label separately? Thank you.