import copy import gradio as gr from gradio_molecule3d import Molecule3D import Bio import Bio.SeqUtils from utils.util_functions import merge_ranges from predict import model_predict from constants import * def update_reps_based_on_radio(*args): struct, text = args[0], args[1] background, model, active_sites = args[2:4], args[4], args[5:] predicted_sites, confs, sequence = model_predict(model, struct, text) merged_sites = merge_ranges(predicted_sites, max_value=len(sequence)) confidence_details = [] new_reps = [] # 1. cal summary summary_text = [] for k, v in predicted_sites.items(): if len(v) > 0: summary_text.append(f"{len(v)} {no_cat_dict[k]} site(s)") if len(summary_text) == 0: summary_text = ["No active sites identified."] summary_text = '; '.join(summary_text) # 2. cal dataframe detail_predicted_sites = {'b':[], '0':[], '1':[], '2':[], '3':[], '4':[], '5':[]} ass = [] for k, v in predicted_sites.items(): for vv in v: detail_predicted_sites[k].append( {'residue_type': sequence[vv-1], 'number': vv, 'confidence': confs[vv-1]} ) ass.append(vv) for i in range(len(sequence)): if i+1 not in ass: detail_predicted_sites['b'].append( {'residue_type': sequence[i], 'number': i+1, 'confidence': confs[i]} ) # 2.1 处理背景 backgrounds = detail_predicted_sites.get('b', []) for r in backgrounds: confidence_details.append([ 'Background', Bio.SeqUtils.seq3(r['residue_type']).upper(), r['number'], r.get('confidence', 'N/A') ]) # 2.2 处理活性位点 for i in range(0, len(active_sites), 2): x, y = active_sites[i], active_sites[i+1] site_key = str(i//2) sites = detail_predicted_sites.get(site_key, []) for s in sites: confidence_details.append([ no_cat_dict[site_key], Bio.SeqUtils.seq3(s['residue_type']).upper(), s['number'], s.get('confidence', 'N/A') ]) # 3. cal reps # 3.1 background ranges = merged_sites['b'] for r in ranges: old_reps = copy.deepcopy(default_reps)[0] old_reps['style'] = background[0][0].lower() + background[0][1:] old_reps['color'] = background[1][0].lower() + background[1][1:] + "Carbon" old_reps['residue_range'] = r new_reps.append(old_reps) # 3.2 active sites for i in range(0, len(active_sites), 2): x, y = active_sites[i], active_sites[i+1] ranges = merged_sites[str(i//2)] for r in ranges: old_reps = copy.deepcopy(default_reps)[0] old_reps['style'] = x[0].lower() + x[1:] old_reps['color'] = y[0].lower() + y[1:] + "Carbon" old_reps['residue_range'] = r new_reps.append(old_reps) return summary_text, confidence_details, Molecule3D(label="Identified Functional Sites", reps=new_reps) def disable_fn(*x): return [gr.update(interactive=False)] * len(x) def able_tip(): return gr.update(visible=True) def check_input(input): if input is not None: return gr.update(interactive=True) return gr.update(interactive=False) with gr.Blocks(title="M3Site-app", theme=gr.themes.Default()) as demo: gr.Markdown("# M3Site: Leveraging Multi-Class Multi-Modal Learning for Accurate Protein Active Site Identification and Classification") gr.Markdown(""" ## Overview **M3Site** is an advanced tool designed to accurately identify and classify protein active sites using a multi-modal learning approach. By integrating protein sequences, structural data, and functional annotations, M3Site provides comprehensive insights into protein functionality, aiding in drug design, synthetic biology, and understanding protein mechanisms. """) gr.Markdown(""" ## How to Use 1. **Select the Model**: Choose the pre-trained model for site prediction from the dropdown list. 2. **Adjust Visual Settings**: Customize the visual style and color for background and active sites. 3. **Upload Protein Structure**: Provide the 3D structure of the protein. You can upload from local or download from PDB Assym. Unit, PDB BioAssembly, AlphaFold DB, or ESMFold DB. 4. **Enter Function Prompt**: Optionally provide a text description of the protein's function. If unsure, leave it blank. 5. **Click "Predict"**: Hit the 'Predict' button to initiate the prediction. The predicted active sites will be highlighted in the structure visualization. 6. **View Results**: The detailed results will be displayed below, including the identified active sites, their types, and confidence scores. """) with gr.Accordion("General Settings (Set before prediction)"): with gr.Row(): model_drop = gr.Dropdown(model_list, label="Model Selection", value=model_list[0]) gr.Markdown("") gr.Markdown("") with gr.Row(): with gr.Row(): style_dropb = gr.Dropdown(style_list, label="Style (Background)", value=style_list[0], min_width=1) color_dropb = gr.Dropdown(color_list, label="Color (Background)", value=color_list[0], min_width=1) with gr.Row(): style_drop1 = gr.Dropdown(style_list, label="Style (CRI)", value=style_list[1], min_width=1) color_drop1 = gr.Dropdown(color_list, label="Color (CRI)", value=color_list[1], min_width=1) with gr.Row(): style_drop2 = gr.Dropdown(style_list, label="Style (SCI)", value=style_list[1], min_width=1) color_drop2 = gr.Dropdown(color_list, label="Color (SCI)", value=color_list[2], min_width=1) with gr.Row(): style_drop3 = gr.Dropdown(style_list, label="Style (PI)", value=style_list[1], min_width=1) color_drop3 = gr.Dropdown(color_list, label="Color (PI)", value=color_list[3], min_width=1) with gr.Row(): with gr.Row(): style_drop4 = gr.Dropdown(style_list, label="Style (PTCR)", value=style_list[1], min_width=1) color_drop4 = gr.Dropdown(color_list, label="Color (PTCR)", value=color_list[4], min_width=1) with gr.Row(): style_drop5 = gr.Dropdown(style_list, label="Style (IA)", value=style_list[1], min_width=1) color_drop5 = gr.Dropdown(color_list, label="Color (IA)", value=color_list[5], min_width=1) with gr.Row(): style_drop6 = gr.Dropdown(style_list, label="Style (SSA)", value=style_list[1], min_width=1) color_drop6 = gr.Dropdown(color_list, label="Color (SSA)", value=color_list[6], min_width=1) with gr.Row(): gr.Markdown("") gr.Markdown(''' *NOTE:* CRI indicates Covalent Reaction Intermediates, SCI indicates Sulfur-containing Covalent Intermediates, PI indicates Phosphorylated Intermediates, PTCR indicates Proton Transfer & Charge Relay Systems, IA indicates Isomerization Activity, SSA indicates Substrate-specific Activities. ''') with gr.Row(): gr.Markdown("
Input Structure
") gr.Markdown("
Output Predictions
") with gr.Row(equal_height=True): input_struct = Molecule3D(label="Input Protein Structure (Default Style)", reps=reps1) output_struct = Molecule3D(label="Output Protein Structure", reps=[]) with gr.Row(equal_height=True): input_text = gr.Textbox(lines=1, label="Function Prompt", scale=16, min_width=1, placeholder="I don't know the function of this protein.") btn = gr.Button("Predict", variant="primary", scale=1, min_width=1, interactive=False) summary_output = gr.Label(label="", scale=18, min_width=1, show_label=False, elem_classes="info") gr.Markdown("### Result Details") confidence_output = gr.DataFrame(headers=["Active Site Type", "Residue Type", "Residue Number", "Confidence"]) option_list = [ style_dropb, color_dropb, model_drop, style_drop1, color_drop1, style_drop2, color_drop2, style_drop3, color_drop3, style_drop4, color_drop4, style_drop5, color_drop5, style_drop6, color_drop6 ] tips = gr.Markdown("### *Tips: Please refresh the page to make a new prediction.*", visible=False) # gr.Markdown("## Citation") # gr.Markdown("If you find this tool helpful, please consider citing the following papers:") # with gr.Accordion("Citations", open=False): # gr.Markdown('''``` # @inproceedings{ouyangmmsite, # title={MMSite: A Multi-modal Framework for the Identification of Active Sites in Proteins}, # author={Ouyang, Song and Cai, Huiyu and Luo, Yong and Su, Kehua and Zhang, Lefei and Du, Bo}, # booktitle={The Thirty-eighth Annual Conference on Neural Information Processing Systems} # } # @article{ouyangm3site, # title={M3Site: Leveraging Multi-Class Multi-Modal Learning for Accurate Protein Active Site Iden-tification and Classification}, # author={Ouyang, Song and Luo, Yong and Su, Kehua and Zhang, Lefei and Du, Bo}, # journal={xxxx}, # year={xxxx}, # } # ```''') # 绑定事件 input_struct.change(check_input, inputs=input_struct, outputs=btn) btn.click( fn=able_tip, inputs=[], outputs=tips ).then( fn=disable_fn, inputs=option_list, outputs=option_list ).then( fn=update_reps_based_on_radio, inputs=[input_struct, input_text] + option_list, outputs=[summary_output, confidence_output, output_struct] ).then( fn=lambda x: x, inputs=[input_struct], outputs=[output_struct] ) if __name__ == "__main__": demo.launch(share=True, debug=True)