ATLAS integration: RCSB on-demand PDB + pair metadata (query/target conformations)

app/pages/1_🔍_Explorer.py

@@ -14,7 +14,10 @@ from app.utils.data_loader import (
     find_predictions_dir, load_prediction_index, load_modes, load_embeddings,
     load_ground_truth, load_pdb_text, PETIMOT_ROOT
 )
-from app.utils.bio_api import get_protein_mutations, get_sequence_from_pdb, render_sequence_aa
+from app.utils.bio_api import (
+    get_protein_mutations, get_sequence_from_pdb, render_sequence_aa,
+    fetch_pdb_structure, fetch_atlas_pair_info
+)
 from app.components.embedding_viewer import render_embedding_viewer
 from app.components.viewer_3d import render_motion_viewer, render_mode_comparison, render_deformed_viewer, render_animated_viewer, render_pred_vs_gt_viewer
 from app.components.sequence_viewer import render_sequence_viewer, render_displacement_chart
@@ -89,6 +92,33 @@ def render_protein_detail(pred_dir, gt_dir, protein_name, key_suffix="", compact
     if os.path.exists(pdb_path):
         with open(pdb_path) as f:
             pdb_text = f.read()
+    else:
+        # Fallback: download from RCSB on-demand
+        # protein_name is like "1A3RH_5W23J" — use the query part (first)
+        query_part = protein_name.split("_")[0] if "_" in protein_name else protein_name
+        with st.spinner(f"Fetching structure {query_part[:4]} from RCSB..."):
+            pdb_text = fetch_pdb_structure(query_part)
+
+    # Show ATLAS pair metadata
+    if "_" in protein_name:
+        with st.expander("🧬 ATLAS conformational pair", expanded=False):
+            pair_info = fetch_atlas_pair_info(protein_name)
+            if pair_info:
+                _qc, _tc = st.columns(2)
+                q, t = pair_info.get("query", {}), pair_info.get("target", {})
+                with _qc:
+                    st.markdown(f"**Query — {q.get('pdb','')} chain {q.get('chain','')}**")
+                    st.caption(q.get('title','') or 'N/A')
+                    st.caption(f"Resolution: {q.get('resolution','N/A')} Å | {q.get('year','')}")
+                    st.markdown(f"[Open in RCSB 🔗](https://www.rcsb.org/structure/{q.get('pdb','')})",
+                                unsafe_allow_html=False)
+                with _tc:
+                    st.markdown(f"**Target (ground truth) — {t.get('pdb','')} chain {t.get('chain','')}**")
+                    st.caption(t.get('title','') or 'N/A')
+                    st.caption(f"Resolution: {t.get('resolution','N/A')} Å | {t.get('year','')}")
+                    st.markdown(f"[Open in RCSB 🔗](https://www.rcsb.org/structure/{t.get('pdb','')})",
+                                unsafe_allow_html=False)
+
 
     # Controls
     mode_idx = st.slider("Mode", 0, len(modes) - 1, 0, key=f"mode_{key_suffix}")

app/utils/bio_api.py

@@ -187,3 +187,64 @@ def get_protein_mutations(protein_name: str, seq_length: int) -> "np.ndarray | N
         if uid:
             return fetch_mutation_frequency(uid, seq_length)
     return None
+
+
+# ── PDB structure fetching (for 3D viewer & ATLAS pairs) ─────────────
+@st.cache_data(ttl=86400, show_spinner=False)
+def fetch_pdb_structure(protein_name: str) -> "str | None":
+    """
+    Download PDB text for a protein like '1A3RH' from RCSB.
+    Filters to the correct chain. Returns PDB text string or None.
+    """
+    if len(protein_name) < 4:
+        return None
+    pdb_id = protein_name[:4].upper()
+    chain  = protein_name[4].upper() if len(protein_name) >= 5 else None
+    try:
+        r = requests.get(f"https://files.rcsb.org/download/{pdb_id}.pdb", timeout=20)
+        if not r.ok:
+            return None
+        pdb_text = r.text
+        if chain:
+            lines = [l for l in pdb_text.split("\n")
+                     if (l.startswith("ATOM") and len(l) > 21 and l[21] == chain)
+                     or l.startswith(("HEADER", "TITLE", "REMARK", "END"))]
+            pdb_text = "\n".join(lines)
+        return pdb_text
+    except Exception as e:
+        logger.warning(f"PDB download failed for {pdb_id}: {e}")
+        return None
+
+
+@st.cache_data(ttl=86400, show_spinner=False)
+def fetch_atlas_pair_info(pair_name: str) -> dict:
+    """
+    Parse an ATLAS-style pair '1A3RH_5W23J' into two conformations
+    and fetch metadata from RCSB for each.
+    Returns dict with query/target PDB IDs, chains, titles, and resolution.
+    """
+    parts = pair_name.split("_")
+    if len(parts) < 2:
+        return {}
+    query_name, target_name = parts[0], parts[1]
+    query_pdb, query_chain   = query_name[:4].upper(),  (query_name[4].upper()  if len(query_name)  >= 5 else "A")
+    target_pdb, target_chain = target_name[:4].upper(), (target_name[4].upper() if len(target_name) >= 5 else "A")
+
+    result = {
+        "pair":   pair_name,
+        "query":  {"pdb": query_pdb,  "chain": query_chain,  "name": query_name},
+        "target": {"pdb": target_pdb, "chain": target_chain, "name": target_name},
+    }
+    for key, pdb_id in [("query", query_pdb), ("target", target_pdb)]:
+        try:
+            r = requests.get(f"https://data.rcsb.org/rest/v1/core/entry/{pdb_id}", timeout=8)
+            if r.ok:
+                d = r.json()
+                result[key]["title"]      = d.get("struct", {}).get("title", "")[:60]
+                result[key]["resolution"] = (d.get("rcsb_entry_info", {})
+                                              .get("resolution_combined", [None]) or [None])[0]
+                result[key]["method"]     = d.get("rcsb_entry_info", {}).get("experimental_method", "")
+                result[key]["year"]       = d.get("rcsb_accession_info", {}).get("deposit_date", "")[:4]
+        except Exception:
+            pass
+    return result