UMAP App to add (#6)

Same as t-SNE but with UMAP optimized with cosine metric

Co-authored-by: Yojana Gadiya <[email protected]>

pages/5_UMAP_for_smiles_app3.py

@@ -0,0 +1,135 @@
+import streamlit as st
+import pandas as pd
+import numpy as np
+from rdkit import Chem
+from rdkit.Chem import Descriptors, Draw
+from umap import UMAP 
+from bokeh.plotting import figure
+from bokeh.models import ColumnDataSource, HoverTool, CustomJS
+from bokeh.palettes import Category10
+from bokeh.transform import factor_cmap
+import io
+import base64
+
+# Streamlit app title
+st.title("UMAP Plot of Molecules")
+st.markdown("*This Streamlit app* allows user to upload **any** :blue-background[CSV file] containing a column named ***SMILES*** and a label column named ***Type***. It generates a UMAP plot (https://umap-learn.readthedocs.io/) which shows the chemical space calculated on a selection of most common RDKIT 2D descriptors. Plot colors are dependent on ***Type*** column indicating the groups.")
+
+def plot_molecule(smiles):
+    try:
+        mol = Chem.MolFromSmiles(str(smiles))
+        if mol is not None:
+            img = Draw.MolToImage(mol, size=(200, 200))
+            return img
+        else:
+            return None
+    except:
+        return None
+
+def get_molecule_image_src(smiles):
+    img = plot_molecule(smiles)
+    if img:
+        buffered = io.BytesIO()
+        img.save(buffered, format="PNG")
+        img_str = base64.b64encode(buffered.getvalue()).decode()
+        return f"data:image/png;base64,{img_str}"
+    return ""
+
+# File uploader
+uploaded_file = st.file_uploader("Choose a CSV file", type="csv")
+
+if uploaded_file is not None:
+    # Read the CSV file
+    df = pd.read_csv(uploaded_file)
+
+    # Check if required columns exist
+    if 'SMILES' not in df.columns or 'Type' not in df.columns:
+        st.error("The CSV file must contain 'SMILES' and 'Type' columns.")
+    else:
+        # Continue with the analysis
+        st.success("File uploaded successfully. Generating UMAP plot...")
+
+        # Convert SMILES to RDKit molecules
+        mols = [Chem.MolFromSmiles(smiles) for smiles in df['SMILES']]
+
+        # Convert SMILES column to string if it exists
+        if 'SMILES' in df.columns:
+            df['SMILES'] = df['SMILES'].astype(str)
+            df['molecule_img'] = df['SMILES'].apply(get_molecule_image_src)
+
+        # Calculate 2D RDKit descriptors
+        desc_names = ['MolWt', 'FractionCSP3', 'HeavyAtomCount', 'NumAliphaticCarbocycles',
+                      'NumAliphaticHeterocycles', 'NumAliphaticRings', 'NumAromaticCarbocycles',
+                      'NumAromaticHeterocycles', 'NumAromaticRings', 'NumHAcceptors',
+                      'NumHDonors', 'NumHeteroatoms', 'NumRotatableBonds', 'NumSaturatedCarbocycles',
+                      'NumSaturatedHeterocycles', 'NumSaturatedRings', 'RingCount',
+                      'MolLogP', 'MolMR']
+
+        desc_functions = [(name, getattr(Descriptors, name)) for name in desc_names]
+        calc = lambda m: [func(m) for _, func in desc_functions]
+        descriptors = [calc(mol) for mol in mols]
+        descriptors_df = pd.DataFrame(descriptors, columns=desc_names)
+
+        # Perform UMAP
+        umap_model = UMAP(n_components=2, random_state=42, n_neighbors = 15, metric = 'cosine')
+        umap_results = umap_model.fit_transform(descriptors_df)
+
+        # Create a Bokeh ColumnDataSource
+        source = ColumnDataSource(data=dict(
+            x=umap_results[:, 0],
+            y=umap_results[:, 1],
+            smiles=df['SMILES'],
+            type=df['Type'],
+            molecule_img=df['molecule_img']
+        ))
+
+        # Create Bokeh figure
+        p = figure(width=800, height=600, title="UMAP Plot of 2D RDKit Descriptors")
+
+        # Create a color mapper
+        colors = Category10[10][:len(df['Type'].unique())]
+        color_mapper = factor_cmap('type', palette=colors, factors=df['Type'].unique())
+
+        # Create the scatter plot
+        scatter = p.scatter('x', 'y', source=source, color=color_mapper, legend_field='type', size=10, alpha=0.8)
+
+        # Create tooltip HTML
+        tooltip_html = """
+        <div>
+            <div>
+                <img src="@molecule_img" height="200" alt="@molecule_img" width="200">
+            </div>
+            <div>
+                <span style="font-size: 12px; color: #666;">Type: @type</span>
+            </div>
+        </div>
+        """
+
+        hover = HoverTool(renderers=[scatter], tooltips=tooltip_html)
+        p.add_tools(hover)
+
+        # Show the plot in Streamlit
+        st.bokeh_chart(p, use_container_width=True)
+
+        # Function to get SVG download link
+        import tempfile
+        import os
+        from bokeh.io import export_svgs
+
+        def get_svg_download_link(fig):
+            with tempfile.NamedTemporaryFile(suffix=".svg", delete=False) as temp_file:
+                fig.output_backend = "svg"
+                export_svgs(fig, filename=temp_file.name)
+
+            with open(temp_file.name, "rb") as file:
+                svg_content = file.read()
+
+            os.unlink(temp_file.name)  # Delete the temporary file
+
+            b64 = base64.b64encode(svg_content).decode()
+            return f'data:image/svg+xml;base64,{b64}'
+
+        # Add download button
+        if st.button("Download Plot as SVG"):
+            svg_href = get_svg_download_link(p)
+            st.markdown(f'<a href="{svg_href}" download="umap_plot.svg">Download Plot as SVG</a>', unsafe_allow_html=True)