BUG FIX motion metrics plotting full sample

THIS IS A MAJOR BUG FIX the motion metrics were not correctly plotted before this commit the old behaviour: the summary stats were split by group and has a separate calculation for the full sample, and when plotting, the summary stats were the combination of ALL summary stats, rather than just the full sample. This has now been fixed with most of the observation remain the same.
roboneurolibre · Apr 13, 2023 · 4f2548f · 4f2548f
1 parent cadb13d
commit 4f2548f
Show file tree

Hide file tree

Showing 5 changed files with 29 additions and 9 deletions.
diff --git a/fmriprep_denoise/visualization/mean_framewise_displacement.py b/fmriprep_denoise/visualization/mean_framewise_displacement.py
@@ -100,7 +100,9 @@ def plot_stats(stats):
             # rotation=45, ha="right", rotation_mode="anchor"
         )
         ax.set_title(
-            f"{dataset}\nMean\u00B1SD={mean_fd:.2f}\u00B1{sd_fd:.2f}; $N={df.shape[0]}$"
+            f"{dataset}\n"
+            f"Motion in dataset subgroups\n"
+            f"Mean\u00B1SD={mean_fd:.2f}\u00B1{sd_fd:.2f}; $N={df.shape[0]}$"
         )
 
         # statistical annotation

diff --git a/fmriprep_denoise/visualization/motion_metrics.py b/fmriprep_denoise/visualization/motion_metrics.py
@@ -13,14 +13,14 @@
         "label": "Percentage %",
         "title": "Significant QC-FC in connectomes\n"
         + r"(uncorrrected, $\alpha=0.05$)",
-        "ylim": (0, 60),
+        "ylim": (0, 70),
     },
     "fdr_p_values": {
         "var_name": "qcfc_fdr_significant",
         "label": "Percentage %",
         "title": "Significant QC-FC in connectomes\n"
         + r"(FDR corrected, $\alpha=0.05$)",
-        "ylim": (0, 25),
+        "ylim": (0, 35),
     },
     "median": {
         "var_name": "qcfc_mad",
@@ -32,7 +32,7 @@
         "var_name": "corr_motion_distance",
         "label": "Pearson's correlation, absolute value",
         "title": "DM-FC",
-        "ylim": (0, 0.33),
+        "ylim": (0, 0.65),
     },
     "modularity": {
         "var_name": "modularity",
@@ -83,15 +83,20 @@ def plot_stats(data, measure):
         fontsize="x-large",
     )
     for i, dataset in enumerate(data):
+        df = data[dataset].query("groups=='full_sample'")
+        baseline_values = df.query("strategy=='baseline'")
+        baseline_mean = baseline_values[measure["label"]].mean()
         sns.barplot(
             y=measure["label"],
             x="strategy",
-            data=data[dataset],
+            data=df,
             ax=axs[i],
             order=strategy_order,
             ci=95,
             palette=paired_palette,
         )
+        # horizontal line baseline
+        axs[i].axhline(baseline_mean, ls="-.", c=paired_palette[0])
         axs[i].set_title(dataset)
         axs[i].set_ylim(measure["ylim"])
         axs[i].set_xticklabels(
@@ -128,7 +133,6 @@ def plot_joint_scatter(path_root, dataset, fmriprep_version):
     )
     motion = pd.read_csv(path_data, sep="\t", index_col=0)
     data = pd.concat([modularity, motion.loc[modularity.index, :]], axis=1)
-
     data = data.drop("groups", axis=1)
     data.index.name = "participants"
     data = data.reset_index()

diff --git a/fmriprep_denoise/visualization/strategy_ranking.py b/fmriprep_denoise/visualization/strategy_ranking.py
@@ -39,6 +39,7 @@ def load_data(path_root, datasets, criteria_name="stringent"):
                 ascending = m != "modularity"
                 r = (
                     data[d]
+                    .query("groups=='full_sample'")
                     .groupby("strategy")[measure["label"]]
                     .describe()["mean"]
                     .sort_values(ascending=ascending)

diff --git a/fmriprep_denoise/visualization/tables.py b/fmriprep_denoise/visualization/tables.py
@@ -146,8 +146,8 @@ def get_descriptive_data(
     # filter data by proportion vol scrubbed
     if fd_thresh is not None and proportion_thresh is not None:
         scrub_label = (fd2label[fd_thresh], "excised_vol_proportion")
-        keep_scrub = confounds_phenotype[scrub_label] <= proportion_thresh
-        keep_scrub = confounds_phenotype.index[keep_scrub]
+        exclude_scrub = confounds_phenotype[scrub_label] > proportion_thresh
+        keep_scrub = confounds_phenotype.index[~exclude_scrub]
     else:
         keep_scrub = confounds_phenotype.index
     mask_motion = keep_gross_fd.intersection(keep_scrub)

diff --git a/scripts/make_manuscript_figures.py b/scripts/make_manuscript_figures.py
@@ -11,6 +11,7 @@
     degrees_of_freedom_loss,
     mean_framewise_displacement,
     motion_metrics,
+    strategy_ranking,
 )
 
 group_order = {
@@ -100,15 +101,19 @@
         )
         axs_modularity = subfigs_modularity[j].subplots(1, 2, sharey=True)
         for i, dataset in enumerate(data):
+            df = data[dataset].query("groups=='full_sample'")
+            baseline_values = df.query("strategy=='baseline'")
+            baseline_mean = baseline_values[measure["label"]].mean()
             sns.barplot(
                 y=measure["label"],
                 x="strategy",
-                data=data[dataset],
+                data=df,
                 ax=axs_modularity[i],
                 order=strategy_order,
                 ci=95,
                 palette=paired_palette,
             )
+            axs_modularity[i].axhline(baseline_mean, ls="-.", c=paired_palette[0])
             axs_modularity[i].set_title(dataset)
             axs_modularity[i].set_ylim(measure["ylim"])
             axs_modularity[i].set_xticklabels(
@@ -137,3 +142,11 @@
         Path(__file__).parents[1] / "outputs" / "meanfd_modularity.png",
         transparent=True,
     )
+
+    data = strategy_ranking.load_data(path_root, datasets)
+    fig_ranking = strategy_ranking.plot_ranking(data)
+
+    fig_ranking.savefig(
+        Path(__file__).parents[1] / "outputs" / "strategy_ranking.png",
+        transparent=True
+    )