Update README.md

README.md

@@ -30,7 +30,8 @@ This model is trained to predict general binding sites of proteins using on the
 `esm2_t6_8M_UR50D`, trained on [this dataset](https://huggingface.co/datasets/AmelieSchreiber/general_binding_sites). The data is 
 not filtered by family, and thus the model may be overfit to some degree. In the Hugging Face Inference API widget to the right 
 there are three protein sequence examples. The first is a DNA binding protein ([see UniProt entry here](https://www.uniprot.org/uniprotkb/D3ZG52/entry)). 
-Note there is nontrivial (GMPGTGK) overlap in the predicted binding sites and the binding sites given in UniProt. 
+Note there is nontrivial (GMPGTGK) overlap in the predicted binding sites and the binding sites given in UniProt. Note also that 
+some of the extraneous predictions are near misses and are very close to the binding sites given in UniProt. 
 
 The second and third were obtained using [EvoProtGrad](https://github.com/Amelie-Schreiber/sampling_protein_language_models/blob/main/EvoProtGrad_copy.ipynb)
 a Markov Chain Monte Carlo method of (in silico) directed evolution of proteins based on a form of Gibbs sampling. The mutatant-type