update

Dockerfile

@@ -0,0 +1,33 @@
+ARG BASE_IMAGE=mambaorg/micromamba
+ARG BASE_TAG=1.5-jammy
+ARG MAMBA_PYTHON_VERSION=3.10
+
+FROM --platform=linux/amd64 ${BASE_IMAGE}:${BASE_TAG}
+
+ARG MAMBA_PYTHON_VERSION
+ENV DEBIAN_FRONTEND=noninteractive
+
+WORKDIR /usr/src/app
+
+RUN micromamba install -y -n base -c conda-forge \
+    python=${MAMBA_PYTHON_VERSION} \
+    && micromamba clean --all --yes
+
+ARG MAMBA_DOCKERFILE_ACTIVATE=1
+ENV BASH_ENV=/usr/local/bin/_activate_current_env.sh
+ENV LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/opt/conda/lib
+
+RUN pip install --no-cache-dir \
+    "stoic @ git+https://github.com/PickyBinders/stoic.git" \
+    gradio==6.9.0
+
+EXPOSE 7860
+ENV GRADIO_SERVER_NAME="0.0.0.0"
+
+ADD . .
+
+USER $MAMBA_USER
+
+ENTRYPOINT ["/usr/local/bin/_entrypoint.sh"]
+
+CMD ["python", "app.py"]

app.py

@@ -0,0 +1,80 @@
+import time
+
+import torch
+import gradio as gr
+from loguru import logger
+
+from stoic.model import Stoic
+
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+logger.info(f"Loading model on {device}")
+model = Stoic.from_pretrained("PickyBinders/stoic")
+model = model.to(device).eval()
+logger.info("Model loaded")
+
+
+def predict(sequences_text: str, top_n: int) -> tuple[str, str]:
+    sequences = [s.strip() for s in sequences_text.strip().split("\n") if s.strip()]
+    if not sequences:
+        raise gr.Error("Please enter at least one protein sequence.")
+    if len(sequences) > 26:
+        raise gr.Error("Maximum 26 unique chains supported.")
+
+    start = time.time()
+    with torch.no_grad():
+        results = model.predict_stoichiometry(sequences, top_n=top_n)
+    elapsed = time.time() - start
+
+    chain_labels = [chr(ord("A") + i) for i in range(len(sequences))]
+
+    header = "| Rank | " + " | ".join(f"Chain {l}" for l in chain_labels) + " | Stoichiometry |"
+    separator = "|------|" + "|".join("-----" for _ in chain_labels) + "|---------------|"
+    rows = []
+    for rank, candidate in enumerate(results, 1):
+        copies = [candidate.get(seq, 0) for seq in sequences]
+        stoich = "".join(f"{l}<sub>{c}</sub>" for l, c in zip(chain_labels, copies))
+        row = f"| {rank} | " + " | ".join(str(c) for c in copies) + f" | {stoich} |"
+        rows.append(row)
+
+    table = "\n".join([header, separator] + rows)
+
+    legend_lines = ["\n\n**Sequences:**"]
+    for label, seq in zip(chain_labels, sequences):
+        preview = seq[:50] + "..." if len(seq) > 50 else seq
+        legend_lines.append(f"- **Chain {label}**: `{preview}`")
+
+    return table + "\n".join(legend_lines), f"{elapsed:.2f}s"
+
+
+with gr.Blocks(title="Stoic - Protein Stoichiometry Prediction") as app:
+    gr.Markdown(
+        "# Stoic\n"
+        "**Fast and accurate protein stoichiometry prediction**\n\n"
+        "Enter one protein sequence per line (one per unique chain type). "
+        "Stoic predicts how many copies of each chain are present in the assembled complex."
+    )
+
+    with gr.Row():
+        with gr.Column():
+            sequences_input = gr.Textbox(
+                label="Protein Sequences (one per line)",
+                placeholder="MKTLLILTLFLAIAASSASA...\nMGSSHHHHHHSSGLVPR...",
+                lines=6,
+            )
+            top_n = gr.Slider(
+                minimum=1, maximum=10, value=3, step=1,
+                label="Number of candidates to return",
+            )
+            btn = gr.Button("Predict Stoichiometry", variant="primary")
+
+        with gr.Column():
+            results_output = gr.Markdown(value="Results will appear here.")
+            run_time = gr.Textbox(label="Runtime")
+
+    btn.click(
+        predict,
+        inputs=[sequences_input, top_n],
+        outputs=[results_output, run_time],
+    )
+
+app.launch()