app.py

import gradio as gr
from transformers import AutoTokenizer, EsmForProteinFolding
from transformers.models.esm.openfold_utils.protein import to_pdb, Protein as OFProtein
from transformers.models.esm.openfold_utils.feats import atom14_to_atom37
import torch
from logging import getLogger

logger = getLogger(__name__)

def convert_outputs_to_pdb(outputs):
    final_atom_positions = atom14_to_atom37(outputs["positions"][-1], outputs)
    outputs = {k: v.to("cpu").numpy() for k, v in outputs.items()}
    final_atom_positions = final_atom_positions.cpu().numpy()
    final_atom_mask = outputs["atom37_atom_exists"]
    pdbs = []
    for i in range(outputs["aatype"].shape[0]):
        aa = outputs["aatype"][i]
        pred_pos = final_atom_positions[i]
        mask = final_atom_mask[i]
        resid = outputs["residue_index"][i] + 1
        pred = OFProtein(
            aatype=aa,
            atom_positions=pred_pos,
            atom_mask=mask,
            residue_index=resid,
            b_factors=outputs["plddt"][i],
            chain_index=outputs["chain_index"][i] if "chain_index" in outputs else None,
        )
        pdbs.append(to_pdb(pred))
    return pdbs

def fold_prot_locally(sequence):
    logger.info("Folding: " + sequence)
    tokenized_input = tokenizer([sequence], return_tensors="pt", add_special_tokens=False)['input_ids'].cuda()

    with torch.no_grad():
        output = model(tokenized_input)
    pdb = convert_outputs_to_pdb(output)
    return pdb

sample_code = """
## Sample usage

    from gradio_client import Client
    
    client = Client("https://wwydmanski-esmfold.hf.space/")
    
    def fold_huggingface(sequence, fname=None):
        result = client.predict(
                    sequence,	# str in 'sequence' Textbox component
                    api_name="/predict")
        result = eval(result)[0]
        if fname is None:
            with tempfile.NamedTemporaryFile("w", delete=False, suffix=".pdb", prefix="esmfold_") as fp:
                fp.write(result)
                fp.flush()
                return fp.name
        else:
            with open(fname, "w") as fp:
                fp.write(result)
                fp.flush()
            return fname
    
    pdb_fname = fold_huggingface("MALWMRLLPLLALLALWGPDPAAAFVNQHLCGSHLVEALYLVCGERGFFYTPKTRREAEDLQVGQVELGGGPGAGSLQPLALEGSLQKRGIVEQCCTSICSLYQLENYCN")

"""

tokenizer = AutoTokenizer.from_pretrained("facebook/esmfold_v1")
model = EsmForProteinFolding.from_pretrained("facebook/esmfold_v1", low_cpu_mem_usage=True).cuda()
model.esm = model.esm.half()
torch.backends.cuda.matmul.allow_tf32 = True

iface = gr.Interface(fn=fold_prot_locally, inputs="text", outputs="text", article=sample_code, title="ESMFold")

iface.launch()