Update app.py

app.py

@@ -19,14 +19,10 @@ DEVICE        = "cuda" if torch.cuda.is_available() else "cpu"
 RNG_SEED = int(os.getenv("RNG_SEED", "42"))
 
 # WT definition (used for parity ELBO computation)
-WT_ID       = "OR293627.1|translated"
 WT_SEQUENCE = (
     "MLGKCLTAGYCSQLLFFWCIVPFCFAALVNAASNSSSHLQLIYNLTICELNGTDWLNQKFDWAVETFVIFPVLTHIVSYGALTTSHFLDTAGLITVSTAGYYHGRYVLSSIYAVFALAALICFVIRLTKNCMSWRYSCTRYTNFLLDTKGNLYRWRSPVVIERRGKVEVGDHLIDLKRVVLDGSAATPITKISAEQWGRP"
 )
 
-# Fitness rescale
-MIN_F, MAX_F = -6200.0, 6200.0
-
 # Valid symbols for ESM
 VALID = set("ACDEFGHIKLMNPQRSTVWYBXZ")
 
@@ -81,12 +77,12 @@ def elbo_batch_from_embeddings(model: VAE_model, emb_batch: torch.Tensor, determ
             ELBO_b, BCE_b, KLD_b = model.all_likelihood_components(emb_batch, deterministic=True)
             return ELBO_b.detach().cpu().numpy()
         except TypeError:
-            # Fallback: use a reseeded single-sample ELBO (still reproducible)
+            # Fallback: use a reseeded single-sample ELBO 
             _set_seeds(RNG_SEED)
             ELBO_b, BCE_b, KLD_b = model.all_likelihood_components(emb_batch)
             return ELBO_b.detach().cpu().numpy()
     else:
-        # Explicitly stochastic mode (not used in this app)
+        # Explicitly stochastic mode 
         ELBO_b, BCE_b, KLD_b = model.all_likelihood_components(emb_batch)
         return ELBO_b.detach().cpu().numpy()
 
@@ -221,8 +217,7 @@ def _ensure_ready():
 
 def score(sequence: str) -> str:
     """
-    Deterministic scoring (Option 2):
-      - Try to use the model's deterministic path (posterior mean z=mu).
+    Try to use the model's deterministic path (posterior mean z=mu).
       - If not supported, fall back to reseeding before each ELBO call
         so identical inputs produce identical ELBOs.
     """
@@ -258,7 +253,7 @@ def score(sequence: str) -> str:
         f"WT ELBO: {wt_elbo:.4f}\n"
         f"Seq ELBO: {test_elbo:.4f}\n"
         f"Evol. index (Seq - WT): {evol_index:.4f}\n"
-        f"Fitness (0–100): {fitness:.2f}"
+        f"Fitness: {fitness:.2f}"
     )
 
 # ---------------------------
@@ -273,7 +268,7 @@ demo = gr.Interface(
         placeholder="Sequence must be exactly 200 aa long"
     ),
     outputs=gr.Textbox(label="Result"),
-    title="Fitness scorer (ESM2 + VAE on embeddings)",
+    title="EscaPRRS Fitness scorer (ESM2 + VAE on embeddings)",
 )
 
 if __name__ == "__main__":