let's try something

app.py

@@ -12,6 +12,11 @@ from run import run, GUIDE_LEN, NUCLEOTIDE_TOKENS
 #     st.write("model prediction: ", returned_x)
 #     return csv_x
 
+@st.cache
+def convert_df(df):
+    # IMPORTANT: Cache the conversion to prevent computation on every rerun
+    return df.to_csv().encode('utf-8')
+
 
 # title and instructions
 st.title('TIGER Cas13 Efficacy Prediction')
@@ -19,7 +24,6 @@ st.session_state['userInput'] = ''
 st.session_state['userInput'] = st.text_input('Enter target transcript (or substring):')
 
 
-predictions = pd.DataFrame.from_dict({'Target Site': [''], 'Normalized LFC': [0.0]})
 if len(st.session_state['userInput']) < GUIDE_LEN:
     st.write('Transcript length must be >= 23 bases. It is {:d} chars'.format(len(st.session_state['userInput'])))
     st.write("")
@@ -27,6 +31,7 @@ elif all([True if nt.upper() in NUCLEOTIDE_TOKENS.keys() else False for nt in st
     st.write('This is your sequence', st.session_state['userInput'])
     predictions = run(st.session_state['userInput'])
     st.write('Model predictions: ', predictions)
+    csv = convert_df(predictions)
+    st.download_button(label='Download CSV file', data=csv, file_name='tiger_predictions.csv', mime='text/csv')
 else:
     st.write('only ACTG is allowed')
-st.download_button(label='Download as CVS File', data=predictions.to_csv())