Tutorials

Interactive notebooks demonstrating MaldiBatchKit workflows end to end. Each notebook is self-contained and can be downloaded from the GitHub repository.

• MaldiBatchKit - Quick Start
  • 1. Load the demo dataset
  • 2. Fit a ComBat corrector
  • 3. Run the diagnostic report
  • 4. Plot the before / after summary
  • Where next
• Comparing correction methods
  • 1. Load the dataset
  • 2. Baseline metrics (no correction)
  • 3. Run every corrector
  • 4. Visual comparison
  • 5. Per-batch peak-position drift, before vs after
• Quality-weighted ComBat - deep dive
  • 1. MALDI-Kleb-AI SNR distribution
  • 2. Fit both ComBat variants
  • 3. Diagnostics side-by-side
  • 4. Effective batch sizes under quality weighting
  • 5. Posterior γ comparison
  • 6. SNR-quantile sensitivity
• Integrating with MaldiAMRKit (MaldiSet)
  • 1. Load the MALDI-Kleb-AI MaldiSet
  • 2. Inspect the cross-centre batch effect
  • 3. Correct via MaldiSetAdapter
  • 4. Verify the correction
  • 5. Swap in a quality-weighted corrector
  • 6. Continue with MaldiAMRKit downstream tools
  • Summary
• Using MaldiBatchKit correctors in a scikit-learn workflow
  • 1. Load the dataset
  • 2. Why fit_transform before splitting is wrong
  • 3. The correct pattern: fit(X_train) + transform(X_test)
  • 4. The recommended pattern: sklearn.Pipeline + cross-validation
  • 5. Hyperparameter search
• Choosing a corrector: AutoCorrector and BatchCorrectionBenchmark
  • 1. Load the dataset
  • 2. AutoCorrector - swap the method as a hyperparameter
  • 3. BatchCorrectionBenchmark - diagnostic comparison
  • 4. Which tool when?

Example Workflows

Single-Call ComBat

from maldibatchkit import ComBat
from maldibatchkit.diagnostics import diagnostic_report

corrector = ComBat(batch=batch, method="johnson")
X_corrected = corrector.fit_transform(X)

report = diagnostic_report(X, X_corrected, batch)
print(report)

Train/Test Safe Pipeline

from sklearn.model_selection import train_test_split
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.ensemble import RandomForestClassifier
from maldibatchkit import SpeciesAwareComBat

X_train, X_test, y_train, y_test = train_test_split(
    X, y, stratify=batch, random_state=42,
)

pipe = Pipeline([
    ("combat", SpeciesAwareComBat(batch=batch, species=species)),
    ("scaler", StandardScaler()),
    ("clf", RandomForestClassifier(n_estimators=200, random_state=42)),
])
pipe.fit(X_train, y_train)
score = pipe.score(X_test, y_test)

Chained Warping + ComBat

from sklearn.pipeline import Pipeline
from maldibatchkit import BatchAwareWarping, ComBat

pipe = Pipeline([
    ("warp", BatchAwareWarping(batch=batch, method="piecewise",
                               n_segments=8, max_shift=10)),
    ("combat", ComBat(batch=batch, method="fortin",
                      discrete_covariates=species)),
])
X_corrected = pipe.fit_transform(X)

MaldiSet Adapter

from maldiamrkit import MaldiSet
from maldibatchkit.integrations import MaldiSetAdapter
from maldibatchkit import QualityWeightedComBat

ds = MaldiSet.from_directory(
    "spectra/", "metadata.csv",
    aggregate_by=dict(antibiotics="Ceftriaxone"),
)
adapter = MaldiSetAdapter(
    batch_column="Batch",
    species_column="Species",
    quality_column="SNR",
)
corrected_ds = adapter.correct(ds, QualityWeightedComBat)