Predict disease phenotype using ceRNA axis
Predict disease phenotype using ceRNA axis
Load the SCZ data
%load_ext autoreload
import scanpy as sc
import anndata as ad
lncRNA_adata = sc.read_h5ad('../../demo/lncRNA_adata.h5ad')
lncRNA_adata.var_names_make_unique()
miRNA_adata = sc.read_h5ad('../../demo/miRNA_adata.h5ad')
miRNA_adata.var_names_make_unique()
mRNA_adata = sc.read_h5ad('../../demo/mRNA_adata.h5ad')
mRNA_adata.var_names_make_unique()
adata = ad.concat([mRNA_adata, miRNA_adata, lncRNA_adata], axis=1)
adata.obs = mRNA_adata.obs
adata.layers['count'] = adata.X
adata
Predict the phenotype from ceRNA axis
%autoreload
import sys
sys.path.append('../../../../1_GREA')
sys.path.append('../../')
from cernatax.cernatax import CERNATAX
# initialize CERNATAX object
cernatax = CERNATAX()
miRNA_list = ['hsa-miR-485-5p']
ceRNA_list = ['ARHGAP8', 'ENST00000522525']
auc_df, stat_df, perf_df, pred_df = cernatax.pred_pheno(adata, miRNA_list, ceRNA_list)
Plot the prediction
import pandas as pd
import matplotlib.pyplot as plt
import numpy as np
# 2) Set up the figure and axes
fig, ax = plt.subplots(figsize=(5, 5))
p1, = ax.plot(auc_df.index, auc_df["ceRNA_+"], '--', color="C0", label="ceRNA_+")
p2, = ax.plot(auc_df.index, auc_df["miRNA_-"], '--', color="C1", label="miRNA_-")
p3, = ax.plot(auc_df.index, auc_df["Combined"], linestyle='None', marker='s', color="C3", label="score")
ax.set_ylim(-0.1, 1.01)
ax.set_ylabel("Score")
ax.set_xticklabels(auc_df.index, rotation=45, ha='right')
# prediction markers at y = –0.05
x = np.arange(len(auc_df))
y_pred = np.full_like(x, -0.05, dtype=float)
# ★ for Case, ● for Con
markers = pred_df.map({"Case":"*", "Con":"o"}).values
pred_color = pred_df.map({"Case":"C3", "Con":"C4"})
colors = pred_color.values
for xi, yi, mk, col in zip(x, y_pred, markers, colors):
ax.scatter([xi], [yi], marker=mk, c=col, s=100, zorder=5)
legend_lines = [p1, p2, p3,
plt.Line2D([0],[0], marker='*', color='w', markerfacecolor='C3', markersize=10, label='pred=Case'),
plt.Line2D([0],[0], marker='o', color='w', markerfacecolor='C4', markersize=6, label='pred=Con'),
]
plt.legend()
fig.savefig('../../demo_out/pred_score.pdf', format='pdf')
We plotted the prediction scores for each sample as follows:
- The blue dashed line represents the predicted probabilities of the ceRNAs (
ARHGAP8andENST00000522525) being highly expressed. - The orange dashed line indicates the predicted probabilities of the miRNA (
hsa-miR-485-5p) being lowly expressed. - The red rectangular dot shows the combined probability of a sample being an SCZ patient (i.e., exhibiting both high ceRNA and low miRNA expression).
At the bottom of the plot, - A red star denotes samples predicted as SCZ cases, - While a purple circle indicates samples predicted as normal controls.
Evaluation - the ROC and PR curves of the predicted combined scores
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from sklearn.metrics import (
roc_curve, auc,
precision_recall_curve, average_precision_score
)
mymap = {'Case': 1, 'Con': 0}
actual = np.array([ mymap[x.split('-')[0]] for x in pred_df.index ])
scores = auc_df['Combined']
fig, ax_roc = plt.subplots(figsize=(5,5))
ax_pr = ax_roc.twinx() # share the same x‐axis
fpr, tpr, _ = roc_curve(actual, scores)
roc_auc = auc(fpr, tpr)
precision, recall, _ = precision_recall_curve(actual, scores)
prc_auc = average_precision_score(actual, scores)
ax_roc.plot(fpr, tpr, label=f"ROC (AUC={roc_auc:.2f})", lw=2, color='C2')
ax_pr.plot(recall, precision, label=f"PR (AP={prc_auc:.2f})", lw=2, color='C4')
ax_roc.set_xlabel("False Positive Rate / Recall", fontsize=12)
ax_roc.set_ylabel("True Positive Rate (ROC)", color='C2', fontsize=12)
ax_pr .set_ylabel("Precision (PR)", color='C4', fontsize=12)
ax_roc.set_ylim(-0.01, 1.01)
ax_pr.set_ylim(-0.01, 1.01)
lines_roc, labels_roc = ax_roc.get_legend_handles_labels()
lines_pr, labels_pr = ax_pr .get_legend_handles_labels()
ax_roc.legend(lines_roc + lines_pr,
labels_roc + labels_pr,
loc='lower right',
fontsize=10)
plt.title("ROC and PR Curves", fontsize=14)
plt.tight_layout()
plt.show()
fig.savefig('../../demo_out/pred_curve.pdf', format='pdf')
The plot displays the Receiver Operating Characteristic (ROC) and Precision-Recall (PR) curves for the predicted combined probabilities of each sample being SCZ or not, compared to the ground truth labels. The green curve represents the ROC, with an area under the curve (AUC) of 0.95, while the purple curve shows the PR curve, with an average precision (AP) of 0.94. Both metrics indicate that the mRNA-miRNA-lncRNA triplet (ARHGAP8, hsa-miR-485-5p, ENST00000522525) has strong discriminative power as a marker for SCZ.
Evaluation - the F1-score, accuracy, recall and precision of the predicted phenotype label (case vs. control)
import pandas as pd
from sklearn.metrics import f1_score, accuracy_score, recall_score, precision_score
mymap = {'Case': 1, 'Con':0}
actual = [mymap[x.split('-')[0]] for x in pred_df.index]
res = []
pred = pred_df.apply(lambda x: mymap[x])
acc = accuracy_score(actual, pred)
f1 = f1_score(actual, pred)
recall = recall_score(actual, pred)
precision = precision_score(actual, pred)
res.append([acc, f1, recall, precision])
df = pd.DataFrame(res, columns=['ACC', 'F1-Score', 'Recall', 'Precision']).T
df = df.reset_index()
df.columns = ['metric', 'value']
print(pred_df)
df
The results show that the predicted labels (case or control) of the mRNA–miRNA–lncRNA triplet (ARHGAP8, hsa-miR-485-5p, ENST00000522525) differ from the ground truth for only one sample Case-A53.
Plot the evaluation metrics
import matplotlib.pyplot as plt
fig, ax = plt.subplots(figsize=(5,5))
scores, values = df.metric, df.value
bars = ax.bar(scores, values, color=['C0','C1','C2','C3'])
for bar, val in zip(bars, values):
ax.text(
bar.get_x() + bar.get_width()/2,
val - 0.05,
f"{val:.2f}",
ha="center",
va="bottom",
fontsize=11
)
ax.set_ylim(-0.01, 1.01)
ax.set_title("Prediction Metrics", fontsize=11)
plt.tight_layout()
plt.show()
fig.savefig('../../demo_out/pred_metric.pdf', format='pdf')
We evaluated the performance of the mRNA–miRNA–lncRNA triplet (ARHGAP8, hsa-miR-485-5p, ENST00000522525) by comparing its predicted labels (case or control) against the ground truth for each sample. The results show an accuracy (ACC) of 0.91, F1-score of 0.91, recall of 0.83, and precision of 1. These metrics further demonstrate that the ceRNA triplet possesses strong discriminative power as a marker for SCZ.