upate

README.md

@@ -1,6 +1,6 @@
 ![banner](assets/FBMN-STATS-GUIed_logo2.png)
 
-[![Open in Statistics App!](https://static.streamlit.io/badges/streamlit_badge_black_white.svg)](https://metabolomics-statistics.streamlit.app/)
+[![Open in Statistics App!](https://static.streamlit.io/badges/streamlit_badge_black_white.svg)](https://fbmn-stats.streamlit.app/)
 
 A web app implementation of the [statistics notebooks](https://github.com/Functional-Metabolomics-Lab/Statistical-analysis-of-non-targeted-LC-MSMS-data) for metabolomics by the [Functional Metabolomics Lab](https://github.com/Functional-Metabolomics-Lab). These notebooks are developed by the Virtual Multi Omics Lab ([VMOL](https://vmol.org/)).
 

pages/4_Hierarchical_Clustering_&_Heatmap.py

@@ -9,7 +9,7 @@
 with st.expander("📖 About"):
     st.markdown(
         """
-Hierarchical clustering analysis (HCA) is a popular unsupervised machine learning technique used for grouping data points based on their similarities. In this method, the data is organized in a tree-like structure or dendrogram, where each branch represents a cluster of data points with similar features. The clustering process starts with each data point being considered as a separate cluster, and then iteratively combines similar clusters until all the data points are in a single cluster.
+Hierarchical clustering analysis (HCA) is a popular unsupervised technique used for grouping data points based on their similarities. In this method, the data is organized in a tree-like structure or dendrogram, where each branch represents a cluster of data points with similar features. The clustering process starts with each data point being considered as a separate cluster, and then iteratively combines similar clusters until all the data points are in a single cluster.
 
 The clustering relies on a distance matrix 'distm', calculated from the feature quantification table (submitted during the data preparation stage), using a specific distance metric  (e.g., Euclidean, Canberra). By default, we use Euclidean distance. Another important factor is the linkage method, which measures the distance between these clusters (e.g., complete, single, average). Our default choice uses the 'complete' method, it calculates the maximum distance between clusters before merging them. Currently, users do not have the option to adjust the distance metric or linkage method. Following HCA, a dendrogram is produced, showing the distances (or 'heights') at which clusters merge or split along the y-axis. The dendrogram offers insights into the clustering process and the relationships between data points. There are a lot of [good videos](https://www.youtube.com/watch?v=7xHsRkOdVwo) and resources out there explaining very well the principle behind clustering. 
 

pages/6_Parametric_assumptions_evaluation.py

@@ -33,20 +33,20 @@
         help="Select two options.",
     )
     if st.session_state.test_attribute and len(st.session_state.test_options) == 2:
-        tabs = st.tabs(["📊 Normal distribution", "📊 Equal variance"])
+        tabs = st.tabs(["📊 Normal distribution (Shapiro-Wilk test)", "📊 Equal variance (Levene test)"])
         with tabs[0]:
-            fig = test_normal_distribution(st.session_state.test_attribute, st.session_state.test_options)
+            fig = test_normal_distribution(st.session_state.test_attribute, st.session_state.test_options, corrections_map[st.session_state.p_value_correction])
             if fig:
                 show_fig(fig, "test-normal-distribution")
         with tabs[1]:
-            fig = test_equal_variance(st.session_state.test_attribute, st.session_state.test_options)
+            fig = test_equal_variance(st.session_state.test_attribute, st.session_state.test_options, corrections_map[st.session_state.p_value_correction])
             show_fig(fig, "test-equal-variance")
 
     st.info(
         """💡 **Interpretation**
 
 In both tests low p-values indicate that data points for a feature are **NOT** normal distributed or have similar variance.
-To meet **parametric** criteria the p-values in the histograms should be equally distributed between 0 and 1.
+To meet **parametric** criteria the p-values in the histograms should not be smaller than 0.05.
 When a larger number of data points indicate low p-values, it would be advisable to opt for a **non-parametric** statistical test.
 """
     )

src/cleanup.py

@@ -203,7 +203,7 @@ def normalization(feature_df, meta_data_df, normalization_method):
     #     normalized = PQN_normalization(feature_df ,ref_norm = "median" , verbose=False)
 
     elif normalization_method == "Total Ion Current (TIC) or sample-centric normalization":
-        normalized = feature_df.apply(lambda x: x/np.sum(x), axis=0)
+        normalized = feature_df.apply(lambda x: x/np.sum(x), axis=1)
 
     else:
         return md_samples, feature_df

src/testparametric.py

@@ -2,15 +2,16 @@
 import pandas as pd
 import plotly.express as px
 import scipy.stats as stats
+import pingouin as pg
 
 
 @st.cache_data
-def test_equal_variance(attribute, between):
+def test_equal_variance(attribute, between, correction):
     # test for equal variance
     data = pd.concat([st.session_state.data, st.session_state.md], axis=1)
     variance = pd.DataFrame(
         {
-            f"{between[0]} - {between[1]}": [
+            f"{between[0]} - {between[1]}": pg.multicomp([
                 stats.levene(
                     data.loc[
                         (data[attribute] == between[0]),
@@ -22,19 +23,19 @@ def test_equal_variance(attribute, between):
                     ],
                 )[1]
                 for f in st.session_state.data.columns
-            ]
+            ], method=correction)[1]
         }
     )
     fig = px.histogram(
         variance,
-        nbins=100,
+        nbins=20,
         template="plotly_white",
+        range_x=[-0.025, 1.025],
     )
-    fig.update_traces(marker_color="#696880")
     fig.update_layout(
         bargap=0.2,
         font={"color": "grey", "size": 12, "family": "Sans"},
-        title={"text": f"TEST FOR EQUAL VARIANCE", "font_color": "#3E3D53"},
+        title={"text": f"TEST FOR EQUAL VARIANCE (LEVENE)", "font_color": "#3E3D53"},
         xaxis_title="p-value",
         yaxis_title="count",
         showlegend=False
@@ -43,7 +44,7 @@ def test_equal_variance(attribute, between):
 
 
 @st.cache_data
-def test_normal_distribution(attribute, between):
+def test_normal_distribution(attribute, between, correction):
     # test for normal distribution
     data = pd.concat([st.session_state.data, st.session_state.md], axis=1)
     for b in between:
@@ -52,34 +53,33 @@ def test_normal_distribution(attribute, between):
             return None
     normality = pd.DataFrame(
         {
-            f"{b}": [
+            f"{b}": pg.multicomp([
                 stats.shapiro(
                     data.loc[
-                        (data[attribute] == between[0]),
+                        (data[attribute] == b),
                         f,
                     ]
                 )[1]
                 for f in st.session_state.data.columns
-            ]
+            ], method = correction)[1]
             for b in between
         }
     )
 
     fig = px.histogram(
-        normality.iloc[:, 1],
-        nbins=100,
+        normality,
+        nbins=20,
         template="plotly_white",
-        color_discrete_sequence=["#696880", "#ef553b"],
-        opacity=0.8,
+        range_x=[-0.025, 1.025],
+        barmode="group",
     )
-    fig.update_traces(marker_color="#696880")
 
     fig.update_layout(
         bargap=0.2,
         font={"color": "grey", "size": 12, "family": "Sans"},
-        title={"text": f"TEST FOR NORMALITY", "font_color": "#3E3D53"},
+        title={"text": f"TEST FOR NORMALITY (SHAPIRO-WILK)", "font_color": "#3E3D53"},
         xaxis_title="p-value",
         yaxis_title="count",
-        showlegend=False
+        showlegend=True
     )
     return fig