Upload folder using huggingface_hub

app.py

@@ -661,9 +661,9 @@ def proteins_tab():
 def clustering_tab():
     return dbc.Container([
         dbc.Row([
-            make_dropdown("Model", "dd-model", MODELS, value="cambridgeltl__SapBERT-from-PubMedBERT-fulltext" if "cambridgeltl__SapBERT-from-PubMedBERT-fulltext" in MODELS else (MODELS[0] if MODELS else None)),
+            make_dropdown("Model", "dd-model", MODELS, value="kamalkraj__BioSimCSE-BioLinkBERT-BASE" if "kamalkraj__BioSimCSE-BioLinkBERT-BASE" in MODELS else (MODELS[0] if MODELS else None)),
             make_dropdown("Min community size", "dd-min", MINS, value="2" if "2" in MINS else (MINS[0] if MINS else None)),
-            make_dropdown("Threshold", "dd-threshold", THRESHOLDS, value="0.8" if "0.8" in THRESHOLDS else (THRESHOLDS[0] if THRESHOLDS else None)),
+            make_dropdown("Threshold", "dd-threshold", THRESHOLDS, value="0.9" if "0.9" in THRESHOLDS else (THRESHOLDS[0] if THRESHOLDS else None)),
             make_dropdown("Field", "dd-field", CLUSTER_FIELDS, value="organism_source" if "organism_source" in CLUSTER_FIELDS else (CLUSTER_FIELDS[0] if CLUSTER_FIELDS else None)),
         ], className="mb-3 g-3"),
         dbc.Row([
@@ -766,17 +766,25 @@ _METRIC_LABELS = {
 }
 
 
-def _build_grid_summary_figure():
-    """Heatmap of mean silhouette (all fields × all min-sizes) per (model, threshold)."""
+def _build_grid_summary_figure(field=None):
+    """Heatmap of mean silhouette per (model, threshold).
+    field=None or '__global__' → average over all fields and min-sizes.
+    field='elution_buffer' etc. → restrict to that field only.
+    """
     if METRICS_DF.empty:
         return go.Figure()
 
     def _short(m):
         return m.split("__")[-1] if "__" in m else m
 
+    src = METRICS_DF.dropna(subset=["silhouette_cosine"])
+    if field and field != "__global__":
+        src = src[src["field"] == field]
+        if src.empty:
+            return go.Figure()
+
     agg = (
-        METRICS_DF
-        .dropna(subset=["silhouette_cosine"])
+        src
         .groupby(["model", "threshold"])["silhouette_cosine"]
         .mean()
         .reset_index()
@@ -823,10 +831,11 @@ def _build_grid_summary_figure():
         hovertemplate="Model: %{y}<br>Threshold: %{x}<br>Mean silhouette: %{z:.3f}<extra></extra>",
     ))
 
+    field_label = "Global (mean over all fields)" if (not field or field == "__global__") else field.replace("_", " ").title()
     n_models = len(pivot.index)
     fig.update_layout(
         title=dict(
-            text="Grid Search Overview — Mean Silhouette (all fields & min-sizes)  ★ = best",
+            text=f"Grid Search Overview — Mean Silhouette · {field_label}  ★ = best",
             font=dict(size=13),
         ),
         xaxis=dict(title="Threshold", type="category", tickfont=dict(size=11)),
@@ -839,21 +848,39 @@ def _build_grid_summary_figure():
     return fig
 
 
+_SUMMARY_FIELD_OPTIONS = (
+    [{"label": "Global (mean over all fields)", "value": "__global__"}]
+    + [{"label": f.replace("_", " ").title(), "value": f} for f in sorted(METRIC_FIELDS)]
+)
+
+
 def grid_metrics_tab():
     if METRICS_DF.empty:
         return dbc.Container([html.P("No metrics data found.", className="text-muted mt-3")])
 
-    summary_fig = _build_grid_summary_figure()
-
     return dbc.Container([
-        # ── Grid search overview (static, built at startup) ────────────────────
+        # ── Grid search overview (callback-driven) ────────────────────────────
         dbc.Card([
             dbc.CardBody([
-                dcc.Graph(
-                    figure=summary_fig,
-                    config={"displayModeBar": True,
-                            "toImageButtonOptions": {"format": "svg", "filename": "grid_search_summary"}},
-                    style={"height": f"{summary_fig.layout.height}px"},
+                dbc.Row([
+                    dbc.Col([
+                        html.Label("View by field", className="fw-semibold small mb-1"),
+                        dcc.Dropdown(
+                            id="dd-summary-field",
+                            options=_SUMMARY_FIELD_OPTIONS,
+                            value="__global__",
+                            clearable=False,
+                            style={"fontSize": "13px"},
+                        ),
+                    ], width=4),
+                ], className="mb-2"),
+                dcc.Loading(
+                    dcc.Graph(
+                        id="metrics-summary-graph",
+                        config={"displayModeBar": True,
+                                "toImageButtonOptions": {"format": "svg", "filename": "grid_search_summary"}},
+                    ),
+                    type="circle", color="#1a73e8",
                 ),
             ], className="p-2"),
         ], className="mb-3 shadow-sm border-0"),
@@ -1379,19 +1406,24 @@ def readme_tab():
 
             # ── Clustering Explorer ───────────────────────��───────────────────
             dbc.AccordionItem(title="Clustering Explorer", children=[
-                html.P(
+                html.P([
                     "Explore semantic clusters of extracted field values. Embeddings are computed "
                     "with biomedical language models; community detection groups semantically similar "
                     "entries into clusters. All plots are rendered natively in Plotly — hover, zoom, "
-                    "and pan are fully interactive.",
+                    "and pan are fully interactive. ",
+                    html.Span("Recommended settings: ", className="fw-semibold"),
+                    "Model = BioSimCSE-BioLinkBERT-BASE, Threshold = 0.9, Min community size = 2 "
+                    "(best mean silhouette across all fields).",
+                ],
                     className="mb-3",
                 ),
                 html.H6("Parameter dropdowns", className="fw-bold"),
                 dbc.ListGroup([
                     dbc.ListGroupItem([
                         html.Span("Model  ", className="fw-semibold"),
-                        "Sentence-embedding model used to encode field values "
-                        "(e.g. PubMedBERT, BioBERT, SapBERT).",
+                        "Sentence-embedding model used to encode field values. "
+                        "BioSimCSE-BioLinkBERT-BASE achieves the highest mean silhouette (0.899) "
+                        "and is selected by default.",
                     ]),
                     dbc.ListGroupItem([
                         html.Span("Min community size  ", className="fw-semibold"),
@@ -1399,7 +1431,8 @@ def readme_tab():
                     ]),
                     dbc.ListGroupItem([
                         html.Span("Threshold  ", className="fw-semibold"),
-                        "Cosine similarity threshold for cluster membership. Higher = tighter clusters.",
+                        "Cosine similarity threshold for cluster membership. Higher = tighter clusters. "
+                        "0.9 is the recommended value for BioSimCSE.",
                     ]),
                     dbc.ListGroupItem([
                         html.Span("Field  ", className="fw-semibold"),
@@ -1426,6 +1459,15 @@ def readme_tab():
                         "common cluster labels for the selected cross-filter field and current "
                         "model / threshold settings.",
                     ]),
+                    dbc.ListGroupItem([
+                        html.Span("Filter by publication era  ", className="fw-semibold"),
+                        "Restrict all clusters to papers published in a specific time window "
+                        "(approximated from PMID ranges: pre-2001 through 2021–2023). "
+                        "Leave blank to include all time periods. "
+                        "Use this to explore how purification practices evolved over time — "
+                        "for example, set Field = Induction Temperature and step through eras "
+                        "to see the shift from 37 °C to low-temperature induction.",
+                    ]),
                 ], flush=True, className="mb-3"),
                 html.H6("UMAP cluster plot", className="fw-bold mt-2"),
                 dbc.ListGroup([
@@ -1466,11 +1508,22 @@ def readme_tab():
             dbc.AccordionItem(title="Grid Metrics", children=[
                 html.P(
                     "Interactive charts comparing clustering quality across all combinations of "
-                    "embedding model, similarity threshold, and min community size. "
+                    "embedding model, similarity threshold, and min community size "
+                    "(14 models × 5 thresholds × 5 min-sizes). "
                     "Loaded from the pre-computed metrics CSVs at startup.",
                     className="mb-3",
                 ),
-                html.H6("Controls", className="fw-bold"),
+                html.H6("Grid Search Overview heatmap", className="fw-bold"),
+                html.P([
+                    "The top heatmap shows mean silhouette score for every (model, threshold) pair. "
+                    "Rows are sorted by best score; the best cell is marked ★. "
+                    "Use the ",
+                    html.Span("View by field  ", className="fw-semibold"),
+                    "dropdown to switch between a global view (mean over all fields and min-sizes) "
+                    "and any individual extraction field (e.g. Elution Buffer, Induction Temperature). "
+                    "The figure can be downloaded as SVG via the camera icon.",
+                ], className="mb-3"),
+                html.H6("Detailed view controls", className="fw-bold"),
                 dbc.ListGroup([
                     dbc.ListGroupItem([
                         html.Span("Field  ", className="fw-semibold"),
@@ -1488,21 +1541,20 @@ def readme_tab():
                         ], className="mb-0 mt-1"),
                     ]),
                     dbc.ListGroupItem([
-                        html.Span("Model  ", className="fw-semibold"),
-                        "Filter the model-comparison bar chart to a single embedding model, "
-                        "or select \"All\" to compare all models side by side.",
+                        html.Span("Model (heatmap)  ", className="fw-semibold"),
+                        "Select the embedding model shown in the threshold × min-size heatmap below.",
                     ]),
                 ], flush=True, className="mb-3"),
-                html.H6("Charts", className="fw-bold mt-2"),
+                html.H6("Detailed charts", className="fw-bold mt-2"),
                 dbc.ListGroup([
                     dbc.ListGroupItem([
-                        html.Span("Heatmap  ", className="fw-semibold"),
-                        "Threshold (x) vs min community size (y) coloured by the selected metric. "
-                        "Hover to see exact values.",
+                        html.Span("Threshold × min-size heatmap  ", className="fw-semibold"),
+                        "For the selected model and field: threshold (x) vs min community size (y) "
+                        "coloured by the chosen metric. Hover to see exact values.",
                     ]),
                     dbc.ListGroupItem([
                         html.Span("Model comparison bar chart  ", className="fw-semibold"),
-                        "Average metric value per model across all parameter combinations.",
+                        "Best metric value per model for the selected field, across all parameter combos.",
                     ]),
                     dbc.ListGroupItem([
                         html.Span("Top-10 configurations table  ", className="fw-semibold"),
@@ -2502,6 +2554,14 @@ def download_cluster_table(n_clicks, rows):
     return dcc.send_data_frame(pd.DataFrame(rows).to_csv, "cluster_proteins.csv", index=False)
 
 
+@app.callback(
+    Output("metrics-summary-graph", "figure"),
+    Input("dd-summary-field", "value"),
+)
+def update_summary_graph(field):
+    return _build_grid_summary_figure(field)
+
+
 @app.callback(
     Output("metrics-table-container", "children"),
     Input("dd-model", "value"),