MAPLE (Methylation-Anchor Probe for Low-Signal Enrichment) is a technology designed to enhance the detection of low-frequency, disease-specific methylation haplotypes. Traditional hybrid capture methods rely on long, unbiased probes and require extremely high sequencing depth, with most reads being uninformative. MAPLE instead uses rationally designed, ultra-sensitive short probes to selectively enrich these rare methylation haplotypes, achieving higher detection sensitivity while substantially reducing sequencing costs.
MAPLE consists of four major components:
The MAPLE probe design module generates highly specific methylation-targeted probes for downstream enrichment and sequencing. It ensures both sequence specificity and thermodynamic stability.
Key Steps:
- Generate haplotype sequences and DNA chains (
MAPLE_sequence_generator.py)
Constructs all possible haplotype sequences and corresponding DNA strands for the target regions. - Evaluate thermodynamic properties and QC (
MAPLE_probe_thermodynamics.py)
Filters probes based on melting temperature, GC content, and secondary structure to ensure optimal hybridization. - Detect off-target binding using BLASTN (
MAPLE_parallel_blastn.py)
Checks potential off-target sites against the genome to minimize non-specific binding. - Select best probes per haplotype (
MAPLE_select_best_probe.py)
Chooses probes with maximal specificity, coverage, and thermodynamic efficiency for each haplotype.
Inputs:
- Target genomic coordinates (start/end positions)
- Methylation haplotype status for the target region
Outputs:
- Qualified probe sequences optimized for specificity and thermodynamic stability
This module processes targeted NGS sequencing data to quantify probe performance and call haplotype-specific methylation patterns.
Key Steps:
- Full NGS pipeline (
MAPLE_NGS.sh)
Includes read trimming, alignment to reference genome, probe evaluation, and quality control. - Count on-target and haplotype fragments (
haplo_fraction.sh)
Calculates the number of captured fragments per probe and haplotype, enabling evaluation of enrichment efficiency. - Compute haplotype frequency ratios (
MAPLE_cal_haplo_fraction.py)
Determines the relative abundance of each haplotype, providing essential input for downstream classification and enrichment factor modeling.
Inputs:
- Paired-end FASTQ files from MAPLE-captured sequencing
Outputs:
- Haplotype fragment counts per target
- Haplotype fraction matrix for downstream analysis
The Enrichment Factor Model quantifies probe enrichment efficiency across target regions. It evaluates how well methylation-specific probes capture the intended genomic loci relative to background noise.
Key Features:
- Calculates on-target coverage for each probe
- Computes haplotype-specific enrichment metrics
- Supports quality control by identifying low-efficiency probes
- Outputs numerical enrichment scores for probe optimization
Inputs:
- BED files of stitched fragments from MAPLE-Stitch
- Reference haplotype methylation pattern definitions
Outputs:
- Per-probe enrichment factors
- Summary statistics for on-target coverage and haplotype specificity
- Intermediate files for QC and downstream analysis
The Boost-tree Classifier uses gradient-boosted trees (LightGBM) to predict cancer or non-cancer status based on enriched methylation haplotypes. It leverages both probe-level enrichment scores and haplotype patterns as features.
Key Features:
- Bayesian optimization of hyperparameters for maximal sensitivity at a given specificity
- Supports stratified or random cross-validation for robust performance evaluation
- Produces per-sample predictions with probability scores and optimal decision thresholds
- Outputs sensitivity, specificity, and ROC-AUC metrics for validation and test datasets
Inputs:
- Probe enrichment scores from EFM
- Sample metadata (e.g., labels, stratification features)
Outputs:
- Trained LightGBM model and optimized hyperparameters
- Predicted probabilities and binary labels for train/validation/test sets
- Performance metrics for evaluation (sensitivity, specificity, AUC)
This project is implemented in Python and requires Python 3.8 or higher. You can set up the project by cloning the repository or downloading it as a ZIP archive.
pandas– Data handling and table operationsnumpy– Numerical computationsBioPython– Sequence handling, melting temperature calculation, BLASTN interfaceBio.SeqUtils.MeltingTempBio.Blast.Applications.NcbiblastnCommandline
primer3– Thermodynamic calculations for oligonucleotidespyfaidx– Fast access to FASTA reference files
Bismark– Bisulfite sequencing alignmentfastp– FASTQ preprocessing and quality controlsamtools– BAM/SAM file processing and indexingMAPLE_stitch– In-house binary to stitch paired-end reads into fragments
pandas– DataFrame manipulation and file I/Onumpy– Numerical computations and array handlingmatplotlib– Plotting and visualizationmatplotlib.backends.backend_pdf.PdfPages– Multi-page PDF output
scipy– Optimization and statistical functionsoptimize– Curve fittingstats.linregress– Linear regression
scikit-learn– Model evaluation utilitiesmetrics.r2_score– Compute R² for model fitting
pandas– Data handling and table operationsnumpy– Numerical computationsscikit-learn– Machine learning utilitiesmetrics– Performance metrics (ROC-AUC, confusion matrix)model_selection– Train/test split, cross-validation
lightgbm– Gradient boosting classifierbayes_opt– Bayesian optimization for hyperparameter tuningjoblib– Model saving and loading
Note: It is recommended to use a Python ≥3.8 environment and install dependencies via
piporconda.
MAPLE enables ultra-sensitive detection of low-frequency cfDNA methylation haplotypes using short capture probes with cost-efficient performance. https://doi.org/10.1101/gr.280736.125
This project is licensed under the GPL-3.0-or-later License - see the LICENSE file for details.