diff --git a/.gitignore b/.gitignore
index beaf9be..b445909 100644
--- a/.gitignore
+++ b/.gitignore
@@ -6,3 +6,4 @@
 *predictions/*
 *visualize.py*
 *seqvec/*
+*data/ESMFold_strcts*
diff --git a/src/ESMfolding.py b/src/ESMfolding.py
new file mode 100644
index 0000000..d16d867
--- /dev/null
+++ b/src/ESMfolding.py
@@ -0,0 +1,72 @@
+import torch
+import os
+from loguru import logger
+from Bio import SeqIO
+from argparse import ArgumentParser
+from transformers import EsmForProteinFolding
+
+
+def read_fasta(fast_file: str):
+    data = {
+        'id': [],
+        'seq': [],
+        'pdb': []
+    }
+
+    with open(fast_file) as handle:
+        for record in SeqIO.parse(handle, "fasta"):
+            data['id'].append(str(record.id))
+            data['seq'].append(str(record.seq))
+    return data
+
+
+def esm2fold(fasta_file: str,
+             result_file: str,
+             device: torch.device = torch.device('cpu'),
+             batch_size: int = 16):
+
+    my_data = read_fasta(fasta_file)
+    esmfold = EsmForProteinFolding.from_pretrained("facebook/esmfold_v1")
+    esmfold.to(device)
+    esmfold.eval()
+
+    ind = 0
+    n_seqs = len(my_data['seq'])
+
+    while ind < n_seqs:
+
+        logger.info(f'index: {ind}')
+
+        if ind + batch_size > n_seqs:
+            batch_size = n_seqs - ind
+
+        with torch.no_grad():
+            my_data['pdb'] += esmfold.infer_pdbs(my_data['seq'][ind:ind + batch_size])
+
+        ind += batch_size
+
+    for i in range(len(my_data['id'])):
+        with open(os.path.join(result_file, f"{my_data['id'][i]}.pdb"), 'w') as f:
+            f.write(my_data['pdb'][i])
+            
+    return
+
+
+if __name__ == "__main__":
+
+    parser = ArgumentParser(description='ESM2 folding')
+    # data
+    parser.add_argument('-seqs', type=str, required=True, 
+                        help='.fasta file of sequences')
+
+    parser.add_argument('-out', type=str, required=True,
+                        help='pdbs saving directory')
+
+    args = parser.parse_args()
+    device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
+
+    esm2fold(
+        fasta_file=args.seqs,
+        result_file=args.out,
+        device=device
+    )
\ No newline at end of file