added app.py

app.py

@@ -0,0 +1,52 @@
+import gradio as gr
+import torch
+import numpy as np
+import matplotlib.pyplot as plt
+from enformer_pytorch import Enformer, load_pretrained_from_url
+from einops import rearrange
+
+# Load pretrained Enformer model (or use from_pretrained if you're using HF model)
+model = load_pretrained_from_url("https://dl.fbaipublicfiles.com/enformer/enformer_pytorch.pt")
+model.eval()
+
+# Helper: one-hot encode DNA (A, C, G, T)
+def one_hot_encode(sequence, length=196_608):
+    mapping = {'A': 0, 'C': 1, 'G': 2, 'T': 3}
+    one_hot = np.zeros((length, 4), dtype=np.float32)
+    sequence = sequence.upper().replace("N", "A")  # replace ambiguous bases
+    for i, base in enumerate(sequence[:length]):
+        if base in mapping:
+            one_hot[i, mapping[base]] = 1.0
+    return one_hot
+
+# Prediction function
+def predict_expression(dna_sequence):
+    encoded = one_hot_encode(dna_sequence)
+    input_tensor = torch.tensor(encoded).unsqueeze(0)  # (1, length, 4)
+    input_tensor = rearrange(input_tensor, 'b l c -> b c l')  # (1, 4, length)
+    
+    with torch.no_grad():
+        output = model(input_tensor)
+        expression = output['human']  # shape: (1, 896, 5313)
+        avg_expr = expression[0].mean(dim=0).numpy()  # average across sequence positions
+
+    # Plot first 10 tissues (customize as needed)
+    plt.figure(figsize=(12, 4))
+    plt.bar(range(10), avg_expr[:10])
+    plt.xticks(range(10), [f"Tissue {i}" for i in range(10)])
+    plt.ylabel("Predicted Expression")
+    plt.title("Gene Expression Prediction (avg across bins)")
+    plt.tight_layout()
+
+    return plt.gcf()
+
+# Gradio Interface
+demo = gr.Interface(
+    fn=predict_expression,
+    inputs=gr.Textbox(lines=5, label="Paste DNA Sequence (A/C/G/T only, ~200kb)"),
+    outputs=gr.Plot(label="Predicted Gene Expression"),
+    title="Gene Expression Predictor (Enformer)",
+    description="Paste a DNA sequence to predict tissue-specific gene expression using a pretrained Enformer model."
+)
+
+demo.launch()