Skip to content

SKlearn-enabled machine learning using ssPA scores

We will use a simple example to demonstrate how to use the sspa scores as features in a machine learning model. We will use ssPA scores generated using the sspa.sspa_SVD (PLAGE) method as input to a logistic regression model to predict COVID-19 status.

Import required packages: 

from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
from sklearn.metrics import roc_auc_score, accuracy_score
from sklearn.metrics import confusion_matrix
Convert COVID status to binary labels 
labels = [1 if i == "COVID19 " else 0 for i in covid_data["Group"]]

Split the data into training and test sets

X_train, X_test, y_train, y_test = train_test_split(covid_data.iloc[:, :-2], labels, test_size=0.2, random_state=42)

Standardise the data

scaler = StandardScaler().fit(X_train)
X_train = pd.DataFrame(scaler.transform(X_train), columns=X_train.columns, index=X_train.index)
X_test = pd.DataFrame(scaler.transform(X_test), columns=X_test.columns, index=X_test.index)

Create ssPA scores for test and train data separately 

sspa_svd_transformer = sspa.sspa_SVD(reactome_pathways, min_entity=3).fit(X_train)
sspa_train = sspa_svd_transformer.transform(X_train)
sspa_test = sspa_svd_transformer.transform(X_test)

We can now fit the model based on pathway scores 

# Create a logistic regression model
clf = LogisticRegression(penalty='l1', solver='liblinear', random_state=42, max_iter=1000)
clf.fit(sspa_train, y_train)

# Predict the test set
y_pred = clf.predict(sspa_test)
And evaluate the model performance

# Calculate the accuracy
print("Accuracy:", accuracy_score(y_test, y_pred))

# Calculate the AUC
print("AUC:", roc_auc_score(y_test, y_pred))

# Plot the confusion matrix
cm = confusion_matrix(y_test, y_pred)
sns.set_style('white')
sns.heatmap(cm, annot=True, fmt='g', cmap='Blues', cbar=False, square=True)
plt.xlabel('Predicted')
plt.ylabel('True')
plt.title('Confusion matrix')
plt.show()
confusion matrix

We can also interpret the model in terms of pathways by looking at feature importance. In this case we will look at the regression coefficients of the logistic regression model.

# Exract the coefficients from the model
coef = pd.DataFrame(clf.coef_.T, index=sspa_train.columns, columns=["Coefficient"])
coef["Pathway"] = coef.index.map(dict(zip(reactome_pathways.index, reactome_pathways['Pathway_name'])))
coef = coef.sort_values(by="Coefficient")

# filter out non-zero coefficients
coef_filt = coef[coef["Coefficient"] != 0]

# Plot the coefficients
plt.figure(figsize=(8, 6))
sns.barplot(data=coef_filt, x="Coefficient", y="Pathway")
plt.tight_layout()
plt.show()
confusion matrix

Using ssPA functions as part of SKLearn pipelines

We may want to apply an ssPA transformation as part of a pre-processing pipeline to prepare test-train splits. We can integrate ssPA functions into sklearn pipelines like so: 

# create pipeline for missing value imputation, standardisation, and sspa transformation
from sklearn.pipeline import Pipeline
from sklearn.impute import SimpleImputer

# create pipeline
sspa_pipeline = Pipeline([
    ('imputer', SimpleImputer(strategy='median').set_output(transform="pandas")),
    ('scaler', StandardScaler().set_output(transform="pandas")),
    ('sspa', sspa_KPCA(reactome_pathways, min_entity=5))
])

# apply the pipeline to test and train data
sspa_train = sspa_pipeline.fit_transform(X_train)
sspa_test = sspa_pipeline.transform(X_test)