Documentation Analysis Methods Molecular Data Processing
Browse Documentation
Molecular Data Processing QTL and GWAS Mapping Integrated Trait–Gene Analysis Methods

Analysis Methods

Molecular Data Processing

Follow the analysis workflow from sequencing reads and genotypes to FPKM, PSI and PDUI

Workflow and dataset context

The analysis scripts cover genotype preparation, RNA-seq processing, splicing, APA, molecular QTL mapping, GWAS, integrated analyses and variant annotation. The settings below describe the accompanying workflow. Imported datasets can differ in cohort, reference and historical processing; use the selected result's metadata and threshold when interpreting a website result.

Genotypes and RNA-seq

Genotype preparation
WGS and RNA-seq have separate alignment and variant-calling routes. Cohort processing includes joint genotyping, source-specific merging where applicable, and genotype QC. The common QC template retains variants with MAF ≥ 0.05 and missingness ≤ 10%; study-specific preparation steps can differ.
RNA-seq alignment
fastp cleans reads, HISAT2 aligns them to the study's analysis reference, and SAMtools sorts and indexes the alignments. Intron-length settings depend on the species; strand settings must match the library protocol.
Expression quantification
StringTie quantifies reference-guided transcripts and genes. The matrix scripts produce FPKM and TPM tables and estimated count tables. The website's expression panel uses FPKM; TPM and coverage-derived estimated counts are distinct units.

Splicing and APA quantification

MAJIQ and VOILA
MAJIQ builds splice graphs and estimates sample-level PSI. VOILA classifies events and exports junction measurements. The PSI matrix retains local-splicing-variation, event, junction and coordinate identity; a single event can contain multiple component measurements.
DaPars2
Coverage profiles and annotated analysis regions are used to infer proximal poly(A) sites and fit short and long 3′ UTR expression. PDUI is the estimated long-isoform fraction, L / (L + S), on a 0–1 scale. The workflow sets Coverage_threshold = 20. Low coverage or a failed estimate can leave a missing value; missing is not zero usage.
Region and site identity
APA matrices retain gene, analysis interval and inferred proximal-site identity. Dataset-specific fits can infer different sites within the same interval. Proximal and distal follow transcription direction, including on the negative strand. The distal reference boundary is not an independently predicted cleavage site.

Prepare phenotypes for association testing

Sample identifiers and reference coordinates are aligned before association testing. The workflow retains filtered molecular measurements separately from the transformed association inputs; the values shown in the molecular-phenotype plots keep their original FPKM, PSI or PDUI units.

Missing values
At most 20% missing measurements per feature are allowed before association testing. Missing values are not treated as zero expression or zero usage.
Expression filtering
Mean FPKM > 0.1, with FPKM > 0.1 in at least 10% of observed samples.
PSI and PDUI filtering
At least 10% of observed values within 0.1–0.9, range ≥ 0.1 and sample variance ≥ 0.001. These are feature-variability filters, not tests for differential splicing or APA.
Transformation
QTL and TWAS inputs use a rank-based inverse normal transformation for each feature. EWAS uses the filtered molecular matrix before that transformation, with samples matched separately for each trait.
Covariates
The molecular QTL template combines five genotype principal components with 20 PEER factors estimated from log2(FPKM + 1). The script parameters and actual analysis logs determine the covariates for a specific run.

Reproduce an analysis

  1. Match the inputsUse the same genome assembly, annotation, chromosome naming and sample identifiers. A filename or gene symbol alone does not establish reference compatibility.
  2. Prepare the environmentInstall the tools called by the selected workflow and replace example paths and sample IDs. Run shell scripts from the repository root so relative paths to shared Python and R scripts resolve correctly.
  3. Record the configurationKeep the code version, software versions, parameter values and input sample lists. Replace example paths in the scripts with validated inputs; the code repository does not substitute for study-specific source data.
  4. Retain provenanceSave run logs, full results, filtered results and QC status together. For EWAS, use the actual MOMENT or MOA model recorded in the OSCA log; a .moment filename can contain an MOA fallback result.

Workflow source code

These links point to the reviewed code version f2bde42 (8 November 2026). Dataset-specific metadata and run logs remain the source for historical analysis settings.