diff --git a/README.md b/README.md
index d3ebd99..c73354b 100644
--- a/README.md
+++ b/README.md
@@ -47,7 +47,7 @@ We also have the following files, that allow you to run a demo of the generatati
 
 We also trained an idpGAN version on Cα traces extracted from all-atom simulations performed using the [ABSINTH implicit solvent model](https://pubmed.ncbi.nlm.nih.gov/18506808/), which was found to [reproduce with good accuracy the experimental behavior of several IDPs](https://pubmed.ncbi.nlm.nih.gov/29805999/). ABSINTH simulations can be performed using the [CAMPARI software package](https://campari.sourceforge.net/V4/index.html). In the `data` directory of this repository, we have the following files for this version of idpGAN:
  - `idpgan_training_set.fasta`: idpGAN training set sequences. For this version of the model, we used simulation data from all sequences with length <= 40 residues.
- - `abstest.fasta`: this is the test set sequences that we used for evaluating this version of idpGAN. They are 15 sequences which were originally simulated in [[Mao et al., 2010]](https://pubmed.ncbi.nlm.nih.gov/20404210/). See the [idpGAN article](https://www.biorxiv.org/content/10.1101/2022.06.18.496675v1) for more information.
+ - `abstest.fasta`: this is the test set sequences that we used for evaluating this version of idpGAN. They are 15 sequences which were originally simulated with the ABSINTH potential in [[Mao et al., 2010]](https://pubmed.ncbi.nlm.nih.gov/20404210/). See the [idpGAN article](https://www.biorxiv.org/content/10.1101/2022.06.18.496675v1) for more information.
  - `abs_generator.pt`: PyTorch weights for a generator model pre-trained on ABSINTH simulation data.
  - `abs_selector.pt`: PyTorch weights for a model for selecting the correct mirror images of xyz conformations generated by idpGAN.