Update app.py

app.py

@@ -32,7 +32,7 @@ model_promoter = AutoModelForSequenceClassification.from_pretrained("llm_promote
 model_promoter.eval()
 
 # D. Multi-Feature LightGBM 
-lgbm_path = hf_hub_download(repo_id="Geonomic/Genomic-Oracle-Weights", filename="dnabert_lightgbm_model_feature_type.pkl")
+lgbm_path = hf_hub_download(repo_id="Geonomic/Genomic-Oracle-Weights", filename="dnabert_lightgbm_model_feature_type_v2.pkl")
 raw_lgbm = joblib.load(lgbm_path)
 
 # If it's a dictionary, print the keys to the log and try to extract the model
@@ -57,8 +57,12 @@ model_pheno = BertForSequenceClassification.from_pretrained("Geonomic/Genomic-Or
 model_pheno.eval()
 
 FEATURE_DICT = {
-    0: "CDS (Coding Sequence)", 1: "Intron", 2: "5' UTR",
-    3: "3' UTR", 4: "Promoter", 5: "Enhancer", 6: "lncRNA"
+    0: "Gene/Transcript (Coding/mRNA)",
+    1: "Regulatory Region (Promoter/Enhancer/Silencer)",
+    2: "Long Non-Coding RNA (lncRNA)",
+    3: "Small/Transfer RNA (snRNA/miRNA/tRNA)",
+    4: "Repeat Region / Mobile Genetic Element",
+    5: "Pseudogene"
 }
 
 # ==============================================
@@ -115,7 +119,7 @@ def run_deep_learning_cascade(dna_sequence):
             raw_scores["Phenotype (Lean)"] = prob_lean
 
     # BRANCH B: Promoter Validation (Triggered if Non-Coding AND is Promoter/Enhancer)
-    elif not is_coding and lgb_prediction in [4, 5]:
+    elif not is_coding and lgb_prediction == 1:
         inputs_promo = tokenizer_promoter([clean_seq], return_tensors="pt", max_length=300, truncation=True, padding=True).to(device)
         
         with torch.no_grad():