import gradio as gr import urllib import re import sys import warnings import torch import torch.nn as nn import ipywidgets as widgets from ipywidgets import interact, fixed from utils.helpers import * from utils.voxelization import processStructures from utils.model import Model import numpy as np import os import moleculekit print(moleculekit.__version__) def update(inp, file, mode): try: pdb_file = file.name except: print("using pdbfile") try: pdb_file = inp if ( re.match( "[OPQ][0-9][A-Z0-9]{3}[0-9]|[A-NR-Z][0-9]([A-Z][A-Z0-9]{2}[0-9]){1,2}", pdb_file, ).group() == pdb_file ): urllib.request.urlretrieve( f"https://alphafold.ebi.ac.uk/files/AF-{pdb_file}-F1-model_v2.pdb", f"files/{pdb_file}.pdb", ) except AttributeError: if len(inp) == 4: pdb_file = inp urllib.request.urlretrieve( f"http://files.rcsb.org/download/{pdb_file.lower()}.pdb1", f"files/{pdb_file}.pdb", ) else: return "pdb code must be 4 letters or Uniprot code does not match", "" if mode == "All residues": ids = get_all_protein_resids( f"files/{pdb_file}.pdb", ) else: ids = get_all_metalbinding_resids(f"files/{pdb_file}.pdb") voxels, prot_centers, prot_N, prots = processStructures(pdb_file, ids) device = torch.device("cuda" if torch.cuda.is_available() else "cpu") voxels.to(device) print(voxels.shape) model = Model() model.to(device) model.load_state_dict(torch.load("weights/metal_0.5A_v3_d0.2_16Abox.pth", map_location=torch.device('cpu'))) model.eval() with warnings.catch_warnings(): warnings.filterwarnings("ignore") output = model(voxels) print(output.shape) prot_v = np.vstack(prot_centers) output_v = output.flatten().cpu().detach().numpy() bb = get_bb(prot_v) gridres = 0.5 grid, box_N = create_grid_fromBB(bb, voxelSize=gridres) probability_values = get_probability_mean(grid, prot_v, output_v) print(probability_values.shape) write_cubefile( bb, probability_values, box_N, outname=f"output/metal_{pdb_file}.cube", gridres=gridres, ) message = find_unique_sites( probability_values, grid, writeprobes=True, probefile=f"output/probes_{pdb_file}.pdb", threshold=7, p=0.15, ) return message, molecule( f"files/{pdb_file}.pdb", f"output/probes_{pdb_file}.pdb", f"output/metal_{pdb_file}.cube", ) def test(): x = """
""" return f"""""" def read_mol(molpath): with open(molpath, "r") as fp: lines = fp.readlines() mol = "" for l in lines: mol += l return mol def molecule(pdb, probes, cube): mol = read_mol(pdb) probes = read_mol(probes) cubefile = read_mol(cube) x = ( """
Isovalue 0.5
""" ) return f"""""" metal3d = gr.Blocks() with metal3d: gr.Markdown("# Metal3D") with gr.Group(): inp = gr.Textbox( placeholder="PDB Code or Uniprot identifier or upload file below", label="Input molecule" ) file = gr.File(file_count="single", type="file") with gr.Row(): mode = gr.Radio( ["All metalbinding residues (ASP, CYS, GLU, HIS)", "All residues"], label="Residues to use for prediction", ) custom_resids = gr.Textbox(placeholder="Comma separated list of residues", label="Custom residues") btn = gr.Button("Run") gr.Markdown( """ Inference using CPU-only, can be quite slow for more than 20 residues. Use Colab notebook for GPU acceleration """ ) gr.Markdown("# Output") out = gr.Textbox(label="status") mol = gr.HTML() btn.click(fn=update, inputs=[inp, file, mode], outputs=[out, mol]) metal3d.launch()