Whole-genome bisulfite sequencing data analysis learning module on Google Cloud Platform

Yujia Qin, Angela Maggio, Dale Hawkins, Laura Beaudry, Allen Kim, Daniel Pan, Ting Gong, Yuanyuan Fu, Hua Yang, Youping DengBriefings in Bioinformatics 2024 · DOI

TL;DR

This paper introduces a comprehensive, cloud-based learning module designed to teach researchers how to analyze whole-genome bisulfite sequencing (WGBS) data using Google Cloud Platform (GCP). It addresses the growing need for accessible, scalable, and cost-effective tools for epigenomic data analysis, particularly for those without extensive computational resources.

What Problem Does This Paper Address?

Whole-genome bisulfite sequencing (WGBS) is a powerful technique for understanding DNA methylation, a crucial epigenetic mechanism involved in gene regulation and various diseases. However, analyzing WGBS data is computationally intensive, requiring significant processing power, storage, and specialized bioinformatics expertise. This creates a barrier for many researchers, especially those in under-resourced institutions. Furthermore, the complexity of the analysis pipeline and the need for high-performance computing infrastructure make it challenging to learn and implement effectively. This paper aims to bridge this gap by providing an interactive, step-by-step learning module that leverages the scalability and accessibility of cloud computing, specifically Google Cloud Platform.

Methods

The core of this work is a modular learning platform built on Google Cloud Platform (GCP). The module is structured into four interconnected submodules, each focusing on a specific aspect of WGBS data analysis:

  • Submodule 1 (Bismark): Introduces fundamental WGBS data preprocessing, including quality control, alignment, and methylation calling using the Bismark tool. This submodule uses a small, illustrative dataset for guided learning.
  • Submodule 2 (metilene): Focuses on downstream analysis for identifying differentially methylated regions (DMRs) using the metilene software. It covers both de novo and feature-based DMR detection.
  • Submodule 3 (nf-core/methylseq): Streamlines the preprocessing workflow by introducing the nf-core/methylseq pipeline, built on Nextflow. This submodule emphasizes reproducibility and scalability.
  • Submodule 4 (nf-core/methylseq with Google Batch): Addresses the challenge of analyzing large-scale WGBS datasets by integrating the nf-core/methylseq pipeline with Google Batch. This submodule demonstrates how to leverage cloud computing for efficient processing of massive datasets.

The module utilizes GCP services like Vertex AI (for interactive notebooks), Cloud Storage (for data storage), and Google Batch (for scalable job execution). It also incorporates educational elements like quizzes and flashcards to aid knowledge retention.

Datasets

  • Small Example Dataset: Used in submodules 1, 2, and 3. This dataset is a subsampled version from the original data with accession number GSE41923 (available on NCBI GEO). It specifically focuses on a region of the mouse genome (chr6:4000000-6,000,000) and includes samples from mouse embryonic stem cells (ESCs) with and without kinase inhibitors (2i).
  • Large-Scale Dataset: Used in submodule 4. This is an unpublished dataset consisting of 12 samples with 24 pair-end FASTQ files, downloaded from the Sequence Read Archive (SRA). This dataset is used to demonstrate the scalability of the nf-core/methylseq pipeline with Google Batch.

Key Insights and Ideas

  • Democratizing Epigenomic Analysis: The module significantly lowers the barrier to entry for WGBS data analysis by providing a structured, cloud-based learning environment, making advanced techniques accessible to a wider range of researchers.
  • Scalability is Key: The integration of Google Batch in submodule 4 highlights the critical role of cloud computing in handling the massive data volumes generated by WGBS, enabling analyses that would be prohibitive on local infrastructure.
  • Progressive Learning: The modular design allows users to build their understanding incrementally, starting with fundamental concepts and progressing to more complex and scalable workflows.
  • Reproducibility and Standardization: The use of pipelines like nf-core/methylseq promotes reproducible research by standardizing analysis workflows and managing software dependencies effectively.
  • Cost-Effectiveness through Cloud Optimization: The paper emphasizes that while cloud computing has costs, strategic use of services like Google Batch and customizable machine types can lead to significant cost savings and efficiency gains compared to traditional high-performance computing.

Key Results

  • The learning module successfully guides users through WGBS data preprocessing and DMR identification using established bioinformatics tools.
  • Submodules 1 and 3, using a small dataset on a standard Vertex AI notebook instance, complete processing in approximately 50 minutes. Submodule 2 (metilene) takes 2-3 minutes.
  • Submodule 4, processing a large-scale dataset using Google Batch, completes the analysis in approximately 3 hours. This is a significant speedup compared to running the same dataset on a local high-performance cluster (over 5 days).
  • The module demonstrates the practical application of cloud resources for managing large datasets, showcasing cost-effective optimization strategies.

Limitations and Future Work

  • Dataset Size for Tutorials: While the tutorial datasets are sufficient for learning the workflow, they may not fully showcase the advantages of cloud computing for extremely large-scale analyses. The authors acknowledge that the large-scale submodule uses a moderately sized SRA dataset to manage costs during tutorial execution.
  • Cost Management: For substantial real-world analyses, advanced cost-management strategies and securing funding are essential. The paper mentions that while operational costs for tutorials are low, comprehensive analyses require careful planning.
  • Tool Evolution: The field of bioinformatics is rapidly evolving, with tools and pipelines being continuously updated. The authors note that they used version 1.6.1 of nf-core/methylseq, while the latest version is 2.6.0, indicating the need for ongoing maintenance and updates of the learning module.
  • Future Work: The authors suggest exploring the development of supplementary resources like video tutorials and workshops. They also highlight that expertise gained from this module can be applied to other nf-core pipelines, broadening researchers' analytical horizons across various biological technologies beyond WGBS. The continuous updates and maintenance of the module are crucial for ensuring its continued relevance and effectiveness.

Summary updated

BibTeX
@article{qin2024whole,
  author = {Yujia Qin and Angela Maggio and Dale Hawkins and Laura Beaudry and Allen Kim and Daniel Pan and Ting Gong and Yuanyuan Fu and Hua Yang and Youping Deng},
  title = {{Whole-genome bisulfite sequencing data analysis learning module on Google Cloud Platform}},
  journal = {Briefings in Bioinformatics},
  year = {2024},
  doi = {10.1093/bib/bbae236},
}
Figure