SnakeAltPromoter Facilitates Differential Alternative Promoter Analysis

Jiang Tan, Yuqing Wu, Ruteja Barve, Jared M. Lalmansingh, Fuhai Li, Philip Payne, Nahyun Kong, Sheng Chih Jin, Yuqi Shan, Ruiwen Zhou, Xinzhou Ge, Jingyi Jessica Li, Richard Head, Yidan SunComputational 2026 · DOI

TL;DR

This paper introduces SnakeAltPromoter, a novel, reproducible, and scalable computational workflow for analyzing alternative promoter usage from RNA-seq data. By integrating three established methods and offering benchmarking against CAGE data, it provides researchers with a unified platform to investigate complex gene regulation and isoform diversity.

What Problem Does This Paper Address?

Alternative promoter usage is crucial for generating diverse gene isoforms, impacting development, disease, and cellular processes. However, studying this phenomenon at scale is challenging. While Cap Analysis of Gene Expression (CAGE) provides precise mapping of transcription start sites (TSSs), its high cost limits its application to large cohorts. RNA sequencing (RNA-seq) is a more accessible alternative, but accurately inferring promoter activity from RNA-seq data requires sophisticated computational approaches. Currently, there's a lack of a standardized, reproducible framework that integrates these methods, making it difficult for researchers to leverage existing RNA-seq data for alternative promoter analysis. This gap hinders the full exploration of promoter-level regulation and its implications.

Methods

SnakeAltPromoter is built on the Snakemake workflow management system, ensuring reproducibility and scalability. The workflow comprises two main modules:

  1. Genome Setup: This module prepares essential reference files, including genome indices and a unified promoter-transcript-gene map, ensuring consistency across all analyses.
  2. Alternative Promoter Analysis: This module processes raw RNA-seq data through several steps:
  3. Quality Control and Alignment: Standard tools like FastQC, TrimGalore, and STAR are used for initial data processing and alignment.
  4. Promoter Quantification: Three complementary methods are employed in parallel:
  5. ProActiv: A junction-based approach that quantifies reads spanning the first intron, indicating productive transcription initiation.
  6. Salmon: A transcript-based method that estimates transcript abundance and aggregates promoter-level counts.
  7. DEXSeq: An exon-based method that counts reads in first-exon bins to infer TSS utilization.
  8. Promoter Classification: Promoters are categorized as "major" (most active per gene), "minor/alternative" (other active promoters), or "inactive" based on their quantified activity.
  9. Differential Analysis: DESeq2 is used for differential promoter activity and usage analyses, identifying changes in promoter-level regulation.

A key feature is the benchmarking module, which allows comparison of RNA-seq-derived promoter activity with matched CAGE data, providing a quantitative evaluation of the different RNA-seq methods. The workflow is accessible via both a command-line interface and a user-friendly graphical interface.

Datasets

The study utilized several datasets for evaluation and benchmarking:

  • Human Heart RNA-seq and CAGE data: This dataset, from both healthy individuals and those with heart failure, was used for initial promoter classification, benchmarking against CAGE, and differential promoter activity analysis.
  • Human Brain RNA-seq data: Used to confirm the generalizability of promoter usage trends observed in heart tissue.
  • GM12878 and K562 cell line RNA-seq and CAGE data: Employed to evaluate the robustness of promoter classification and differential analysis across different cell types and to identify cell type-specific promoters.
  • Simulated datasets: Used to assess the impact of sequencing depth and read length on promoter analysis.

The paper does not introduce a new benchmark dataset but rather uses existing CAGE data as a gold standard for evaluating RNA-seq-based methods.

Key Insights and Ideas

  • Complementary Strengths of RNA-seq Methods: No single RNA-seq method is universally superior. ProActiv excels in precision and concordance with CAGE for intron-containing promoters, while Salmon offers higher recall and can detect intronless promoters. DEXSeq provides a flexible exon-binning approach. Integrating these methods within SnakeAltPromoter allows users to leverage their combined strengths.
  • Prevalence of Alternative Promoter Usage: The study confirms that alternative promoter usage is widespread, with a significant proportion of genes utilizing multiple active promoters. This highlights the complexity of gene regulation beyond simple gene-level expression.
  • CAGE as a Robust Reference: CAGE data consistently served as a reliable benchmark, allowing for quantitative evaluation of RNA-seq methods and providing insights into their strengths and limitations.
  • Importance of Reproducibility: The use of Snakemake and Conda environments is crucial for ensuring that promoter analysis results are reproducible across different computational environments and over time.

Key Results

  • Scalability: SnakeAltPromoter can process a 50-million-read RNA-seq sample in just over 2 hours on a 32-core workstation, demonstrating its efficiency for large-scale analyses.
  • Promoter Classification: The study found that major promoters are not always the first annotated TSS, with significant proportions identified downstream. Approximately 25% of expressed genes showed multiple active promoters.
  • Benchmarking with CAGE:
    • ProActiv showed the highest precision (78.6%) for major promoters, while Salmon had the highest recall (73.1%).
  • For differential promoter activity, ProActiv demonstrated the strongest overall agreement with CAGE (Pearson's r = 0.83, CCC = 0.74) for intron-containing genes.
  • Salmon and DEXSeq showed lower concordance with CAGE for intronless promoters, highlighting a limitation for these methods.
  • Robustness: Promoter classification and activity quantification results were generally robust across different datasets (heart, brain, cell lines) and technical variations (sequencing depth, read length), with consistent relative performance patterns among the methods.

Limitations and Future Work

  • Intronless Promoter Detection: ProActiv, due to its junction-based design, cannot detect intronless promoters, which represent a significant portion of the human genome. While Salmon and DEXSeq can detect them, their concordance with CAGE is lower for these loci.
  • Long-Read Sequencing Data: The current workflow is designed for short-read RNA-seq. The authors suggest future work to adapt SnakeAltPromoter for long-read sequencing data (e.g., PacBio, Oxford Nanopore) to improve the detection of low-abundance promoters and address limitations of short reads.
  • Integration with Epigenomic Data: Integrating epigenomic data (e.g., ATAC-seq, ChIP-seq) is proposed as a future direction to enhance promoter activity characterization, validate RNA-seq findings, and improve the analysis of low-expression promoters. This could involve machine learning models to combine multiomics data.
  • Technical Challenges for Long Reads: The authors acknowledge challenges in precisely localizing TSSs with long-read data due to incomplete 5' coverage and high indel error rates, requiring specialized protocols and algorithms.

Summary updated

BibTeX
@inproceedings{tan2026snakealtpromoter,
  author = {Jiang Tan and Yuqing Wu and Ruteja Barve and Jared M. Lalmansingh and Fuhai Li and Philip Payne and Nahyun Kong and Sheng Chih Jin and Yuqi Shan and Ruiwen Zhou and Xinzhou Ge and Jingyi Jessica Li and Richard Head and Yidan Sun},
  title = {{SnakeAltPromoter Facilitates Differential Alternative Promoter Analysis}},
  booktitle = {Computational},
  year = {2026},
  doi = {10.34133/csbj.0033},
}
Figure