Spaces:
Paused
Paused
| import json | |
| import os | |
| import numpy as np | |
| import pandas as pd | |
| import gradio as gr | |
| import matplotlib.pyplot as plt | |
| DATA_DIR = "data" | |
| VARIANTS = ["WT", "var1", "var2", "var3", "var4"] | |
| # SWI weights from the paper context [cite: 52] | |
| swi_weights = {'A': 0.835, 'C': 0.111, 'D': 0.921, 'E': 0.963, 'F': 0.354, | |
| 'G': 0.732, 'H': 0.654, 'I': 0.443, 'K': 0.901, 'L': 0.512, | |
| 'M': 0.456, 'N': 0.812, 'P': 0.678, 'Q': 0.854, 'R': 0.887, | |
| 'S': 0.776, 'T': 0.798, 'V': 0.567, 'W': 0.312, 'Y': 0.412} | |
| sequences_dict = { | |
| "WT": "EVKLVESGGDLVKPGGSLKLSCAASGFTFSSYGMSWVRQTPDKRLEWVATISRGGSYTYYPDSVKGRFTISRDNAKNTLYLQMSSLKSEDTAMYYCARRETYDEKGFAYWGQGTTVTVSSGGGGSGGGGSGGGGSDIELTQSPSSLTVTAGEKVTMSCKSSQSLLNSGNQKNYLTWYQQKPGQPPKLLIYWASTRESGVPDRFTGSGSGRDFTLTISSVQAEDLAVYYCQNDNSHPLTFGAGTKLELKR", | |
| "var1": "EVKLTESGGKTVKPGGSVTLSCQVSGIDFSSYGMSWVRQTPEDKLEWVATISRGGSYTYIPERYKGRVTISKDFSKSTLYLNMKNLKPSDTATYYCAIRETYDEKGFAHFGQGTKLTVTDDGPDPGSPDPSDPSNDIKLTQSPSEITVNEGEKVTITCNSSQSLLNSGNQKNYLTWYKKKPGEPPQLLIYWASTRESGVSSRFVGSGSGTSFTLTISSFKKEDVATYYCQNDNSHPLTFGQGTKLSLKP", | |
| "var2": "EVKLVESGGATVAPGGSVTLSCKASGFDFSSYGMSWVAQTPEGQLRWVATISRGGSYTYVPEAYAGRVTISRDFAASTAYLHMRDLQPSDTATYYCAVRETYDEKGFAHFGQGTQLTVADPGAAPGAPEPALPSDDIQLTQSPSEVTANLGETVTFTCTSSKSLLNSGNQKNYLTWYKQAPGQPPQLLIYWASTREPGVSSRFVGSGSGTNFTLTISSLQESDVATYYCQNDNSHPLTFGSGTKLSLKK", | |
| "var3": "MVQLVESGGATVAPGGSVTLSCKVSGIDFSSYGMSWVAQTPDNQLRWVATISRGGSYTYIPKSYAGRFTISVDHSKSTAYLNMSNLQPSDTATYYCAIRETYDEKGFAHFGQGTQLTVADPGAAPGAPLEDDPSNDIQLTQSPSEVTVNLGEKVTLTCNSSKSLLNSGNQKNYLTWYQQKPGEPPKLLIYWASTREEGVSSRFVGSGSGTSFTLTISSFQEEDVATYYCQNDNSHPLTFGQGTKLTLKP", | |
| "var4": "SVSLTESGGATVAPGGSVTLSCKASGFDFSSYGMSWVAQTPEGRLLWVATISRGGSYTYVPSRYAGRFTISRDDSKSTLYLNMSNLQPSDTATYYCAVRETYDEKGFAHFGEGTQLTVSDAGEAPGGAEPADPSPDIVLTQSPSSVTVNLGEKVTLTCNSSKSLLNSGNQKNYLTWYKQAPGQPPQLLIYWASTRESGVSSRFVGSGSGTSFTLTITSFQAEDVATYYCQNDNSHPLTFGSGTKLSLKP" | |
| } | |
| # Pre-load WT PAE for difference calculations | |
| def get_wt_pae(): | |
| path = os.path.join(DATA_DIR, "WT", "pae.json") | |
| with open(path) as f: | |
| data = json.load(f) | |
| return np.array(data.get("pae", data.get("predicted_aligned_error"))) | |
| WT_PAE_DATA = get_wt_pae() | |
| def calculate_swi(seq): | |
| return round(sum(swi_weights.get(aa, 0.5) for aa in seq) / len(seq), 3) | |
| def calculate_net_charge(seq): | |
| return seq.count('R') + seq.count('K') - seq.count('D') - seq.count('E') | |
| def calculate_interface_pae(pae, h_range=(0, 115), l_range=(140, 250)): | |
| block1 = pae[h_range[0]:h_range[1], l_range[0]:l_range[1]] | |
| block2 = pae[l_range[0]:l_range[1], h_range[0]:h_range[1]] | |
| return round((np.mean(block1) + np.mean(block2)) / 2, 3) | |
| def load_variant_data(variant): | |
| base = os.path.join(DATA_DIR, variant) | |
| with open(os.path.join(base, "scores.json")) as f: | |
| scores = json.load(f) | |
| with open(os.path.join(base, "pae.json")) as f: | |
| pae_data = json.load(f) | |
| pae = np.array(pae_data.get("pae", pae_data.get("predicted_aligned_error"))) | |
| pdb_path = os.path.join(base, "structure.pdb") | |
| avg_plddt = np.mean(scores["plddt"]) if isinstance(scores["plddt"], list) else scores["plddt"] | |
| ptm = np.mean(scores.get("ptm", 0)) if isinstance(scores.get("ptm"), list) else scores.get("ptm", 0) | |
| return avg_plddt, ptm, pae, pdb_path | |
| def view_structure(pdb_path): | |
| with open(pdb_path) as f: | |
| pdb_data = f.read() | |
| viewer_id = "mol_view_" + str(hash(pdb_path)) | |
| html = f""" | |
| <script src="https://3dmol.org/build/3Dmol-min.js"></script> | |
| <div id="{viewer_id}" style="height: 500px; width: 100%;"></div> | |
| <script> | |
| var viewer = $3Dmol.createViewer("{viewer_id}", {{}}); | |
| viewer.addModel(`{pdb_data}`, "pdb"); | |
| viewer.setStyle({{'cartoon': {{'colorscheme': {{'prop':'b','gradient': 'roygb','min':50,'max':90}}}}}}); | |
| viewer.zoomTo(); | |
| viewer.render(); | |
| </script> | |
| """ | |
| return html | |
| def update_variant(variant): | |
| avg_plddt, ptm, pae, pdb_path = load_variant_data(variant) | |
| i_pae = calculate_interface_pae(pae) | |
| seq = sequences_dict.get(variant, "") | |
| swi_val = calculate_swi(seq) | |
| charge_val = calculate_net_charge(seq) | |
| pos82 = seq[81] if len(seq) > 81 else "N/A" | |
| metrics = pd.DataFrame({ | |
| "Metric": ["Avg pLDDT", "pTM", "Interface PAE", "SWI", "Net Charge", "Pos 82 Residue"], | |
| "Value": [round(float(avg_plddt), 2), round(float(ptm), 3), i_pae, swi_val, charge_val, pos82] | |
| }) | |
| # Generate Difference PAE plot | |
| fig, ax = plt.subplots(figsize=(5, 5)) | |
| if variant == "WT": | |
| im = ax.imshow(pae, cmap="viridis", vmin=0, vmax=30) | |
| ax.set_title("WT: Absolute PAE Map") | |
| cbar_label = "Expected Error (Å)" | |
| else: | |
| # Subtract WT PAE to show relative changes | |
| diff_pae = pae - WT_PAE_DATA | |
| im = ax.imshow(diff_pae, cmap="bwr", vmin=-5, vmax=5) | |
| ax.set_title(f"{variant}: ΔPAE (Relative to WT)") | |
| cbar_label = "Error Difference (Å)" | |
| fig.colorbar(im, ax=ax, label=cbar_label) | |
| ax.set_xlabel("Residue Index") | |
| ax.set_ylabel("Residue Index") | |
| structure_html = view_structure(pdb_path) | |
| return metrics, fig, structure_html | |
| with gr.Blocks() as demo: | |
| gr.Markdown( | |
| """ | |
| <center> | |
| <h1>AI-Assisted Intrabody Design Explorer</h1> | |
| <p>This demo visualizes precomputed AlphaFold2 results for WT and ProteinMPNN-designed scFv variants, following the methodology described in the original paper (Galindo et al. (2026))</p> | |
| </center> | |
| """ | |
| ) | |
| variant = gr.Dropdown(VARIANTS, value="WT", label="Select Design Variant") | |
| with gr.Row(): | |
| metrics_table = gr.Dataframe(label="Structural & Solubility Metrics") | |
| with gr.Column(): | |
| pae_plot = gr.Plot(label="PAE Analysis") | |
| gr.Markdown(""" | |
| ### How to read the ΔPAE Map: | |
| * **🟦 Blue Regions**: **Improved stability.** Error is lower than WT. | |
| * **⬜ White Regions**: **No significant change.** Structural confidence is identical to WT. | |
| * **🟥 Red Regions**: **Reduced stability.** Error is higher than WT (potential distortion). | |
| """) | |
| structure_view = gr.HTML(label="3D Structure (Confidence Colored)") | |
| variant.change(update_variant, inputs=variant, outputs=[metrics_table, pae_plot, structure_view]) | |
| demo.load(update_variant, inputs=variant, outputs=[metrics_table, pae_plot, structure_view]) | |
| demo.launch( | |
| auth=(os.environ.get("app_user"), os.environ.get("app_pass")), | |
| ssr_mode=False | |
| ) |