TECHNICAL RESOURCES · 数据分析

Data Analysis

This page covers: Analysis Pipeline Overview · Preprocessing · Differential-Analysis Methods · Functional Enrichment Analysis · Multi-Omics Integration.

Methodology note · content is citable as-is

KEY POINTS

DESeq2 (negative-binomial model)
edgeR (exact test / GLM)
limma-voom (mean–variance modeling)

01Analysis Pipeline Overview

01Data QC
02Preprocessing
03Differential Analysis
04Functional Enrichment
05Multi-Omics Integration
06Visualization

02Preprocessing

High-quality preprocessing is the foundation of reliable analysis results

Data Quality Control (QC)

Raw-data quality assessment—FASTQ QC (FastQC), mapping-rate statistics and duplication rate—ensuring data reliability and downstream accuracy.

Normalization

Removes batch effects and technical bias with mainstream methods (TMM, RLE, Quantile, ComBat), making samples comparable.

Missing-Value Handling

Multiple imputation methods (KNN, MICE) or removal of high-missingness features, ensuring data integrity and robust analysis.

Data Transformation

log2 transformation, Pareto scaling and z-score standardization to meet the distribution requirements of different analysis models.

03Differential-Analysis Methods

Professional differential-analysis strategies tailored to each omics data type

Transcriptomics

DESeq2 (negative-binomial model)
edgeR (exact test / GLM)
limma-voom (mean–variance modeling)

Proteomics

limma (empirical-Bayes modeling)
MSstats (MS-data statistics)
DEP R package (DIA/DDA data)

Metabolomics

t-test / ANOVA parametric tests
PLS-DA multivariate discriminant analysis
VIP > 1 screening of key metabolites

Multivariate Statistics

PCA unsupervised dimensionality reduction
OPLS-DA supervised classification
permutation-test validation

04Functional Enrichment Analysis

In-depth interpretation of the biological functions and pathway significance of differential molecules

Method	Database / tools	Use case	Output
ORA (hypergeometric test)	GO、KEGG、Reactome	differential gene/protein/metabolite lists	significantly enriched pathways (p-value)
GSEA (gene-set enrichment analysis)	MSigDB, custom gene sets	genome-wide ranked data	NES、FDR、Leading Edge
Pathway-topology analysis	KEGG、MetaboAnalyst	metabolomics data	pathway impact score
Network Analysis	STRING、Cytoscape	protein interaction, gene co-expression	network-topology parameters, hub nodes

05Multi-Omics Integration

Joint cross-omics analysis to uncover deeper biological mechanisms

Correlation Integration

Pearson/Spearman cross-omics correlation analysis builds gene–protein–metabolite association networks, revealing multi-layered regulatory relationships.

Multi-Omics Dimensionality Reduction

MOFA (Multi-Omics Factor Analysis), mixOmics DIABLO and sPLS-DA integrate multi-omics data for dimensionality reduction and classification.

Machine-Learning Integration

Random-forest and XGBoost feature selection builds multi-omics predictive models for biomarker discovery and phenotype prediction.

Q&AFAQ

Should I run DESeq2 or limma for differential expression?

For transcriptomics the options listed are DESeq2 (negative-binomial model), edgeR (exact test / GLM) and limma-voom (mean–variance modeling), chosen to match the omics data type.

Can proteomics and metabolomics share the same differential analysis?

No. Proteomics uses limma (empirical-Bayes modeling), MSstats or the DEP R package; metabolomics uses t-test / ANOVA and PLS-DA, with VIP > 1 to screen key metabolites.

What metrics should I read in the enrichment results?

ORA reports significantly enriched pathways by p-value, GSEA reports NES, FDR and Leading Edge, and pathway-topology analysis reports a pathway impact score.

My samples were processed in different batches — can normalization fix that?

Normalization removes batch effects and technical bias with methods such as TMM, RLE, Quantile and ComBat, making samples comparable.

How do I combine transcriptome, proteome and metabolome data?

Pearson/Spearman correlation builds gene–protein–metabolite association networks; MOFA, mixOmics DIABLO and sPLS-DA cover dimensionality reduction and classification; random forest and XGBoost build predictive models.