'C++' Implementations of Functional Enrichment Analysis

Fast implementations of functional enrichment analysis methods using 'C++' via 'Rcpp'. Currently provides Over-Representation Analysis (ORA), Gene Set Enrichment Analysis (GSEA), design-aware GSEA on RNA-seq counts with per-sample gene set scoring, Weighted Enrichment Analysis for ORA and GSEA, Network-based Set Enrichment Analysis (NSEA), multi-layer network-based enrichment, and multi-omics integration workflows. Additional features include early fusion at the feature level, late fusion at the pathway level, multi-omics contribution tracing, topology-aware explanation helpers, Bayesian term selection, and extremely fast Random Walk with Restart (RWR) using 'RcppEigen'. The enrichment methods build on GSEA by Subramanian et al. (2005) , the multilevel strategy derived from 'fgsea' by Korotkevich et al. (2021) , and network-based enrichment ideas described by Glaab et al. (2012) .

enrichit: C++ Implementations of Functional Enrichment Analysis

enrichit provides C++ implementations of functional enrichment analysis methods and S4 result classes used by the clusterProfiler family. It supports ORA, GSEA, design-aware GSEA on RNA-seq counts, weighted enrichment, network-based enrichment, multilayer network workflows, and multi-omics aggregation and contribution analysis.

Installation

You can install the development version of enrichit from GitHub using devtools:

# install.packages("devtools")
devtools::install_github("YuLab-SMU/enrichit")

Main components

enrichit organizes its functions around five components:

• Core enrichment engines: ORA, GSEA, weighted ORA/GSEA, and GSON-aware variants.
• Design-aware and score-level GSEA: dsea() for enrichment on a count matrix plus a design (rank by the requested contrast, then GSEA), and dsea_scores()/ssgsea_scores() for per-sample gene set scores.
• Network-aware enrichment: single-layer nsea() and multi-layer mnsea() workflows based on Random Walk with Restart.
• Multi-omics integration: early fusion at the feature level and late fusion at the pathway level.
• Contribution and topology outputs: contribution tables and topology-aware extraction helpers for visualization in enrichplot.

Implementation and features

• Implementation: Algorithms are implemented in C++ via Rcpp, with sparse network propagation using RcppEigen.
• ORA: standard hypergeometric ORA with optional weighted ORA through Wallenius' noncentral hypergeometric distribution.
• GSEA: multilevel, permutation, and adaptive strategies for ranked enrichment analysis.
• Design-aware GSEA: dsea() takes a count matrix or a DESeqDataSet plus a design from model.matrix() (rownames must match the sample names), variance-stabilises the counts (vst or rlog), ranks genes by the requested contrast, and runs the GSEA engine already in the package. It returns a plain data.frame -- GSEA's table plus padj and contrast -- so it can be filtered and plotted like any other GSEA result. dsea_scores() and ssgsea_scores() are the other half: a gene set x sample score matrix for analyses that need samples rather than one contrast.
• DESeq2 adapters: as_ranked() and as_genes() turn DESeqResults and fitted DESeqDataSet objects into a ranking or a gene vector, so a dsea() analysis can start from an existing DESeq2 result; extra arguments are passed on to DESeq2::results(), so the contrast uses its c("factor", "numerator", "denominator") form. DESeq2 is suggested, not imported.
• GSON support: native ora_gson() and gsea_gson() interfaces for structured gene set collections.
• NSEA: nsea() and nsea_gson() for network-ranked enrichment on a single graph, including mode = "signed" for bidirectional propagation.
• Multi-layer topology fusion: mnsea() and mnsea_gson() for multiplex or heterogeneous network propagation across multiple layers.
• Multi-omics early fusion: aggregate_omics(), harmonize_ids(), and select_features_for_ora() for feature-level integration before enrichment.
• Multi-omics late fusion: aggregate_enrichment() for pathway-level aggregation of multiple enrichment results.
• Contribution tracing: get_omics_contribution(), classify_omics_pattern(), and get_mnsea_contribution() for contribution summaries.
• Topology-aware extraction: extract_mnsea_subnetwork() for pathway-specific node/edge tables that can be passed to downstream visualization packages.
• Bayesian term selection: bayes_enrich() and bayes_summary() for posterior-based term prioritization.

Main APIs

Classical enrichment

• ora(), ora_gson()
• gsea(), gsea_gson()
• gseaScores()

Weighted enrichment

• ora(..., weight = )
• ora_gson(..., weight = )
• gsea(..., weight = )
• gsea_gson(..., weight = )

Design-aware and score-level GSEA

• dsea()
• dsea_scores(), ssgsea_scores()
• as_ranked(), as_genes()

Network-aware enrichment

• prepare_network()
• nsea(), nsea_gson()
• prepare_multilayer_network()
• propagate_multilayer()
• collapse_multilayer_scores()
• mnsea(), mnsea_gson()

Multi-omics integration

• aggregate_omics()
• harmonize_ids()
• select_features_for_ora()
• aggregate_enrichment()

Explanation helpers

• get_omics_contribution()
• classify_omics_pattern()
• get_mnsea_contribution()
• extract_mnsea_subnetwork()

Result Objects

The package returns the following S4 result classes:

• enrichResult for ORA-like workflows
• gseaResult for ranked enrichment workflows
• nseaResult for single-network propagation plus enrichment
• mnseaResult for multi-layer propagation, collapsed scores, and cached explanation tables

dsea() is the exception on the other side: it returns a plain data.frame rather than a gseaResult. dsea_scores() and ssgsea_scores() return a DseaScores object, an S3 list holding the gene set x sample score matrix ($scores) and the metadata it was built with ($meta).

These classes are used across the clusterProfiler family:

• enrichit handles core computation, algorithm implementation, and contribution data preparation
• clusterProfiler provides high-level biological interpretation workflows and general enrichment analysis interfaces
• enrichplot handles visualization
• gson provides a structured gene set resource layer for managing and exchanging gene set collections across the family
• knowledge-base-oriented downstream packages such as DOSE, ReactomePA, meshes, and MicrobiomeProfiler provide domain-specific annotation and interpretation layers