Update app.py

app.py

@@ -5,9 +5,13 @@ import gradio as gr
 from transformers import pipeline
 
 
-# -----------------------------
-# Configuration
-# -----------------------------
+# =========================================================
+# Config
+# =========================================================
+
+LAYOUT = "alt2"  # "alt2" or "alt3"
+ENABLE_FLAIR = False
+DEFAULT_MODEL = "CeLLaTe_contracted_ent_Reinit_LLRD"
 
 RAW_MODEL_REGISTRY = {
     "CeLLaTe_contracted_ent_Reinit_LLRD": "OTAR3088/CeLLaTe_contracted_ent_Reinit_llrd",
@@ -38,22 +42,6 @@ RAW_MODEL_REGISTRY = {
     "CeLLaTe-V1-Model": "OTAR3088/bioformer-cellate_V1",
 }
 
-# Explicit metadata is better than inferring backend from the model name.
-MODEL_REGISTRY = {
-    name: {"repo_id": repo_id, "backend": "hf", "enabled": True}
-    for name, repo_id in RAW_MODEL_REGISTRY.items()
-}
-
-ENABLE_FLAIR = False
-
-AVAILABLE_MODELS = {
-    name: cfg
-    for name, cfg in MODEL_REGISTRY.items()
-    if cfg["enabled"] and (ENABLE_FLAIR or cfg["backend"] != "flair")
-}
-
-DEFAULT_MODEL = "CeLLaTe_contracted_ent_Reinit_LLRD"
-
 EXAMPLES = [
     """Aside from in vivo models, numerous studies investigating bacterial virulence and pathogenesis have also
 employed in vitro cell line models to gain an initial understanding of the intricate host-pathogen interactions.
@@ -64,32 +52,73 @@ HeLa cells, which originate from human cervical epithelial cells, are thus frequ
 A. baumannii frequently infects human epithelial tissues, such as the respiratory system, skin and mucosal linings [15].
 HeLa cells are resilient and readily cultured in vitro, exhibiting a rapid growth rate.
 This ensures the availability of a uniform and consistent cell population for studies, rendering them economical and reliable.""",
-    """Based on the limma R package, a total of 2578 (DEGs 1398 downregulated and 1188 upregulated) were screened out from GEO: GSE225819 data...""",
-    """Given the dynamic nature of T cell activation and the heterogeneity of CD4+ T cells, we mapped gene expression regulation using single-cell transcriptomes spanning four time points of CD4+ T cell activation...""",
-    """To quantitatively characterise the macrophage foetal-like profile observed, we projected adult and foetal macrophages data from the human decidual–placental interface into our data set...""",
-    """Peripheral blood mononuclear cells (PBMCs) were isolated using Ficoll-Paque PLUS (GE Healthcare) density gradient centrifugation...""",
-    """We performed dimensionality reduction and embedding using uniform manifold approximation (UMAP) (ref. 21) (Methods) and observed that cells separated by time point of stimulation...""",
-    """Finally, we leveraged this model to experimentally evaluate genes linked to immune-related phenotypes by GWAS...""",
+
+    """Based on the limma R package, a total of 2578 (DEGs 1398 downregulated and 1188 upregulated) were screened out from GEO: GSE225819 data, including 20 normal samples and 20 GIST samples with liver metastasis (|log2FC| > 1; P < 0.05), suggesting that these DEGs may be involved in liver metastasis in GIST patients. The top 10 upregulated genes were PENK, IGF2, GPR20, CTSL, SCRG1, PNMAL1, NKX3-2, ANO1, PLAT, and BCHE.
+Moreover, we evaluated IGF2 expression in the GIST cell line. By western blotting, expression levels of IGF2 in GIST882, GIST882-R, GIST-T1, and GIST-T1-R were higher than those in normal RGM-1.""",
+
+    """Given the dynamic nature of T cell activation and the heterogeneity of CD4+ T cells, we mapped gene expression regulation using single-cell transcriptomes spanning four time points of CD4+ T cell activation. We reconstructed activation trajectories for naive and memory CD4+ T cells and identified eQTL effects manifesting at different time points and across different subpopulations of cells.""",
+
+    """To quantitatively characterise the macrophage foetal-like profile observed, we projected adult and foetal macrophages data from the human decidual–placental interface into our data set. This unique tissue setting includes both adult/maternal monocyte-derived macrophages and foetal/placental YS-derived macrophages (Hofbauer cells).""",
+
+    """Peripheral blood mononuclear cells (PBMCs) were isolated using Ficoll-Paque PLUS (GE Healthcare) density gradient centrifugation. Naive (CD25− CD45RA+ CD45RO−) and memory (CD25− CD45RA− CD45RO+) CD4+ T cells were isolated from the PBMC fraction using EasySep naive CD4+ T cell isolation kits and memory CD4+ T cell enrichment kits.""",
+
+    """We performed dimensionality reduction and embedding using uniform manifold approximation (UMAP) and observed that cells separated by time point of stimulation, forming a gradual progression from resting to the most activated cell state. This progression was accompanied by changes in activation markers such as CD69 and IL2RA.""",
+
+    """Finally, we leveraged this model to experimentally evaluate genes linked to immune-related phenotypes by GWAS. Interestingly, CRISPR/Cas9-mediated KO of GWAS hits in iPSC-derived macrophages and DCs highlighted their potential role in physiological and pathological cell states of distinct cell types."""
 ]
 
-EXAMPLE_OPTIONS = {
-    f"Example {i + 1}": text.strip()
-    for i, text in enumerate(EXAMPLES)
+
+def build_model_registry():
+    registry = {}
+    for model_name, repo_id in RAW_MODEL_REGISTRY.items():
+        registry[model_name] = {
+            "repo_id": repo_id,
+            "backend": "hf",     # explicit backend metadata
+            "enabled": True,
+            "tag": "",
+            "description": "Description placeholder. Add a short note about what this model predicts and when to use it."
+        }
+
+    # Example override for the default model
+    registry[DEFAULT_MODEL]["tag"] = "best"
+    registry[DEFAULT_MODEL]["description"] = (
+        "Tags 2 entities with CellType and Tissue contracted into a single entity."
+    )
+
+    return registry
+
+
+MODEL_REGISTRY = build_model_registry()
+
+AVAILABLE_MODELS = {
+    name: cfg
+    for name, cfg in MODEL_REGISTRY.items()
+    if cfg["enabled"] and (ENABLE_FLAIR or cfg["backend"] != "flair")
+}
+
+COLOR_MAP = {
+    "CellLine": "#DBEAFE",
+    "CELL_LINE": "#DBEAFE",
+    "CellType": "#DCFCE7",
+    "CELL_TYPE": "#DCFCE7",
+    "Tissue": "#FCE7F3",
+    "TISSUE": "#FCE7F3",
+    "CELL_TISSUE": "#FCE7F3",
 }
 
 GR_THEME = gr.themes.Soft(
     primary_hue="indigo",
     secondary_hue="rose",
-    neutral_hue="gray",
+    neutral_hue="gray"
 )
 
 CSS = """
 .app-shell {
-    max-width: 1250px;
+    max-width: 1320px;
     margin: 0 auto;
 }
 .hero {
-    padding: 0.25rem 0 0.75rem 0;
+    padding: 0.25rem 0 0.8rem 0;
 }
 .hero h1 {
     margin-bottom: 0.2rem;
@@ -104,68 +133,126 @@ CSS = """
     flex-wrap: wrap;
     margin-top: 0.75rem;
 }
-.badge {
+.pill {
     display: inline-block;
-    padding: 0.25rem 0.7rem;
+    padding: 0.18rem 0.65rem;
     border-radius: 999px;
-    border: 1px solid #d6dcff;
+    font-size: 0.84rem;
+    border: 1px solid #d7defb;
     background: #eef2ff;
-    font-size: 0.9rem;
+    color: #334155;
+}
+.pill-best {
+    background: #ecfeef;
+    border-color: #b7efc2;
+    color: #166534;
+    font-weight: 600;
+}
+.info-card {
+    border: 1px solid #e5e7eb;
+    border-radius: 14px;
+    padding: 14px 16px;
+    background: #ffffff;
+    box-shadow: 0 1px 2px rgba(15, 23, 42, 0.04);
+}
+.info-title {
+    font-size: 0.92rem;
+    color: #667085;
+    margin-bottom: 6px;
+}
+.model-name {
+    font-weight: 700;
+    margin-bottom: 6px;
+}
+.model-desc {
+    color: #334155;
+    line-height: 1.45;
+    margin-top: 8px;
+}
+.model-repo {
+    margin-top: 10px;
+    font-size: 0.82rem;
+    color: #64748b;
+    word-break: break-all;
+}
+.legend-card {
+    border: 1px solid #e5e7eb;
+    border-radius: 14px;
+    padding: 12px 14px;
+    background: #ffffff;
+}
+.section-note {
+    color: #64748b;
+    font-size: 0.92rem;
 }
 """
 
 HEADER_HTML = """
 <div class="app-shell hero">
     <h1>🧬 OTAR3088 Biomedical NER Demo</h1>
-    <p>Extract CeLLaTe biomedical entities from abstracts and paragraphs using trained NER models.</p>
+    <p>Extract CeLLaTe biomedical entities from biomedical abstracts and paragraphs.</p>
     <div class="badge-row">
-        <span class="badge">CellLine</span>
-        <span class="badge">CellType</span>
-        <span class="badge">Tissue</span>
+        <span class="pill">CellLine</span>
+        <span class="pill">CellType</span>
+        <span class="pill">Tissue</span>
     </div>
 </div>
 """
 
-ABOUT_MD = """
-This demo showcases CeLLaTe entity extraction models trained to identify biomedical entities relevant to knowledge discovery.
 
-**How to use**
-1. Choose a model.
-2. Paste biomedical text or load an example.
-3. Run extraction.
-4. Review tagged spans and the entity summary.
+# =========================================================
+# Helpers
+# =========================================================
 
-**Note**  
-This space is intended for research demonstration purposes. Models may be updated as development continues.
-"""
+def model_display_label(model_name: str) -> str:
+    tag = AVAILABLE_MODELS[model_name].get("tag", "").strip()
+    return f"{model_name} · {tag}" if tag else model_name
 
-COLOR_MAP = {
-    "CellLine": "#DBEAFE",
-    "CELL_LINE": "#DBEAFE",
-    "CellType": "#DCFCE7",
-    "CELL_TYPE": "#DCFCE7",
-    "Tissue": "#FCE7F3",
-    "TISSUE": "#FCE7F3",
-    "CELL_TISSUE": "#FCE7F3",
-}
 
+MODEL_CHOICES = [
+    (model_display_label(model_name), model_name)
+    for model_name in AVAILABLE_MODELS.keys()
+]
 
-# -----------------------------
-# Model loading and inference
-# -----------------------------
 
-@lru_cache(maxsize=6)
-def load_model(model_name: str):
+def make_example_label(text: str, idx: int, max_chars: int = 85) -> str:
+    one_line = " ".join(text.split())
+    preview = one_line[:max_chars].rstrip()
+    if len(one_line) > max_chars:
+        preview += "..."
+    return f"Example {idx + 1}: {preview}"
+
+
+EXAMPLE_CHOICES = [
+    (make_example_label(text, i), text.strip())
+    for i, text in enumerate(EXAMPLES)
+]
+
+
+def render_model_card(model_name: str) -> str:
     cfg = AVAILABLE_MODELS[model_name]
+    tag = cfg.get("tag", "").strip()
+
+    tag_html = ""
+    if tag:
+        tag_class = "pill pill-best" if tag.lower() == "best" else "pill"
+        tag_html = f'<span class="{tag_class}">{tag}</span>'
+
+    return f"""
+    <div class="info-card">
+        <div class="info-title">Model description</div>
+        <div style="display:flex; align-items:center; gap:8px; flex-wrap:wrap;">
+            <div class="model-name">{model_name}</div>
+            {tag_html}
+        </div>
+        <div class="model-desc">{cfg.get("description", "Description placeholder.")}</div>
+        <div class="model-repo"><strong>Repo:</strong> {cfg["repo_id"]}</div>
+    </div>
+    """
 
-    if cfg["backend"] != "hf":
-        raise ValueError("Flair models are disabled in this demo.")
 
-    return pipeline(
-        "ner",
-        model=cfg["repo_id"],
-        aggregation_strategy="simple",
-    )
+def load_example(example_text):
+    return example_text or ""
 
 
 def normalize_entities(predictions):
@@ -199,6 +286,18 @@ def summarize_entities(entities):
     return sorted(rows, key=lambda row: (-row[1], row[0]))
 
 
+@lru_cache(maxsize=6)
+def load_model(model_name: str):
+    cfg = AVAILABLE_MODELS[model_name]
+    if cfg["backend"] != "hf":
+        raise ValueError("Flair models are disabled in this space.")
+    return pipeline(
+        "ner",
+        model=cfg["repo_id"],
+        aggregation_strategy="simple",
+    )
+
+
 def tagger(text, model_name):
     text = (text or "").strip()
     if not text:
@@ -207,87 +306,232 @@ def tagger(text, model_name):
     ner_pipe = load_model(model_name)
     predictions = ner_pipe(text)
     entities = normalize_entities(predictions)
+    summary = summarize_entities(entities)
 
-    return {"text": text, "entities": entities}, summarize_entities(entities)
+    return {"text": text, "entities": entities}, summary
 
 
-def load_example(example_name):
-    return EXAMPLE_OPTIONS.get(example_name, "")
+def clear_outputs():
+    return "", {"text": "", "entities": []}, []
 
 
-def clear_all():
-    return "", {"text": "", "entities": []}, []
+def model_table_rows():
+    rows = []
+    for name, cfg in AVAILABLE_MODELS.items():
+        rows.append([
+            name,
+            cfg.get("tag", ""),
+            cfg.get("description", ""),
+            cfg["repo_id"]
+        ])
+    return rows
 
 
-# -----------------------------
-# UI
-# -----------------------------
+# =========================================================
+# Alternative 2: Sidebar + analysis canvas
+# =========================================================
 
-def gradio_ui():
+def build_ui_alt2():
     with gr.Blocks(theme=GR_THEME, css=CSS, title="OTAR3088 Biomedical NER Demo") as demo:
         gr.HTML(HEADER_HTML)
 
-        with gr.Accordion("About this demo", open=False):
-            gr.Markdown(ABOUT_MD)
-
         with gr.Row():
-            model_choice = gr.Dropdown(
-                choices=list(AVAILABLE_MODELS.keys()),
-                value=DEFAULT_MODEL,
-                label="Model",
-                scale=4,
-            )
-
-            example_choice = gr.Dropdown(
-                choices=list(EXAMPLE_OPTIONS.keys()),
-                label="Load example",
-                scale=3,
-            )
-
-            load_btn = gr.Button("Load", scale=1)
-            clear_btn = gr.Button("Clear", scale=1)
-            run_btn = gr.Button("Run extraction", variant="primary", scale=1)
-
-        with gr.Row(equal_height=True):
-            with gr.Column(scale=1):
+            with gr.Column(scale=3, min_width=320):
+                model_choice = gr.Dropdown(
+                    choices=MODEL_CHOICES,
+                    value=DEFAULT_MODEL,
+                    label="Model",
+                    info="Choose the NER model to run"
+                )
+
+                model_info = gr.HTML(
+                    value=render_model_card(DEFAULT_MODEL)
+                )
+
+                example_choice = gr.Dropdown(
+                    choices=EXAMPLE_CHOICES,
+                    value=None,
+                    label="Example text",
+                    info="Select a sample to populate the input box"
+                )
+
+                with gr.Row():
+                    run_btn = gr.Button("Run extraction", variant="primary")
+                    clear_btn = gr.Button("Clear")
+
+                gr.HTML("""
+                <div class="legend-card">
+                    <div class="info-title">Supported entities</div>
+                    <div style="display:flex; gap:8px; flex-wrap:wrap; margin-top:6px;">
+                        <span class="pill">CellLine</span>
+                        <span class="pill">CellType</span>
+                        <span class="pill">Tissue</span>
+                    </div>
+                    <div class="section-note" style="margin-top:10px;">
+                        Paste or load biomedical text, then run inference to inspect highlighted entities and summary counts.
+                    </div>
+                </div>
+                """)
+
+            with gr.Column(scale=8):
                 input_text = gr.Textbox(
                     label="Input text",
                     placeholder="Paste a biomedical abstract or paragraph here...",
-                    lines=18,
+                    lines=16
                 )
 
-            with gr.Column(scale=1):
                 output_highlight = gr.HighlightedText(
                     label="Tagged entities",
                     color_map=COLOR_MAP,
-                    show_legend=True,
+                    show_legend=True
                 )
+
                 entity_summary = gr.Dataframe(
                     headers=["Entity type", "Count", "Avg. confidence"],
                     interactive=False,
-                    label="Entity summary",
+                    label="Entity summary"
                 )
 
-        load_btn.click(
+        example_choice.change(
             fn=load_example,
             inputs=example_choice,
-            outputs=input_text,
+            outputs=input_text
+        )
+
+        model_choice.change(
+            fn=render_model_card,
+            inputs=model_choice,
+            outputs=model_info
         )
 
         run_btn.click(
             fn=tagger,
             inputs=[input_text, model_choice],
-            outputs=[output_highlight, entity_summary],
+            outputs=[output_highlight, entity_summary]
         )
 
         clear_btn.click(
-            fn=clear_all,
-            outputs=[input_text, output_highlight, entity_summary],
+            fn=clear_outputs,
+            outputs=[input_text, output_highlight, entity_summary]
         )
 
     return demo
 
 
+# =========================================================
+# Alternative 3: Tabbed interface
+# =========================================================
+
+def build_ui_alt3():
+    with gr.Blocks(theme=GR_THEME, css=CSS, title="OTAR3088 Biomedical NER Demo") as demo:
+        gr.HTML(HEADER_HTML)
+
+        with gr.Tabs():
+            with gr.Tab("Demo"):
+                with gr.Row():
+                    with gr.Column(scale=4, min_width=320):
+                        model_choice = gr.Dropdown(
+                            choices=MODEL_CHOICES,
+                            value=DEFAULT_MODEL,
+                            label="Model",
+                            info="Choose the NER model to run"
+                        )
+
+                        model_info = gr.HTML(
+                            value=render_model_card(DEFAULT_MODEL)
+                        )
+
+                        example_choice = gr.Dropdown(
+                            choices=EXAMPLE_CHOICES,
+                            value=None,
+                            label="Example text",
+                            info="Select a sample to populate the input box"
+                        )
+
+                        with gr.Row():
+                            run_btn = gr.Button("Run extraction", variant="primary")
+                            clear_btn = gr.Button("Clear")
+
+                    with gr.Column(scale=8):
+                        input_text = gr.Textbox(
+                            label="Input text",
+                            placeholder="Paste a biomedical abstract or paragraph here...",
+                            lines=16
+                        )
+
+                        output_highlight = gr.HighlightedText(
+                            label="Tagged entities",
+                            color_map=COLOR_MAP,
+                            show_legend=True
+                        )
+
+                        with gr.Accordion("Entity summary", open=True):
+                            entity_summary = gr.Dataframe(
+                                headers=["Entity type", "Count", "Avg. confidence"],
+                                interactive=False,
+                                label="Summary"
+                            )
+
+                example_choice.change(
+                    fn=load_example,
+                    inputs=example_choice,
+                    outputs=input_text
+                )
+
+                model_choice.change(
+                    fn=render_model_card,
+                    inputs=model_choice,
+                    outputs=model_info
+                )
+
+                run_btn.click(
+                    fn=tagger,
+                    inputs=[input_text, model_choice],
+                    outputs=[output_highlight, entity_summary]
+                )
+
+                clear_btn.click(
+                    fn=clear_outputs,
+                    outputs=[input_text, output_highlight, entity_summary]
+                )
+
+            with gr.Tab("Models"):
+                gr.Markdown("### Available models")
+                gr.Dataframe(
+                    value=model_table_rows(),
+                    headers=["Model", "Tag", "Description", "Repo"],
+                    interactive=False,
+                    label="Model catalogue"
+                )
+
+            with gr.Tab("About"):
+                gr.Markdown("""
+### About this demo
+
+This space extracts CeLLaTe biomedical entities from biomedical text.
+
+**Supported entities**
+- CellLine
+- CellType
+- Tissue
+
+**How to use**
+1. Choose a model in the Demo tab.
+2. Select an example or paste your own text.
+3. Run extraction.
+4. Review highlighted spans and the summary table.
+
+**Note**
+Descriptions shown for models are placeholders and can be expanded into richer model cards later.
+""")
+
+    return demo
+
+
+# =========================================================
+# Launch
+# =========================================================
+
 if __name__ == "__main__":
-    app = gradio_ui()
+    app = build_ui_alt2() if LAYOUT == "alt2" else build_ui_alt3()
     app.launch()