ARACRA: Automated RNA-seq Analysis for Chemical Risk Assessment

ARACRA is a fully automated, interactive web-based pipeline built on Nextflow and Streamlit that streamlines RNA-seq analysis from raw FASTQ files to transcriptomics Point of Departure (tPoD) calculation, featuring a human-in-the-loop quality control step to facilitate robust chemical risk assessment.

The Gist

Imagine you are a detective trying to solve a mystery: How do different chemicals affect the human body?

In the past, to solve this, scientists had to look at the "instruction manual" inside our cells (our RNA) to see which instructions were being read and which were being ignored when exposed to a chemical. This is called RNA-seq.

However, reading these manuals is incredibly difficult. It's like trying to assemble a 10,000-piece puzzle while wearing thick gloves, using tools from five different hardware stores, and having to write down every single step on a piece of paper before moving to the next tool. Most scientists (especially those who aren't computer experts) get stuck in the middle of the process.

Enter ARACRA.

Think of ARACRA as a high-tech, automated factory that takes a messy pile of raw puzzle pieces and hands you a finished, framed picture, complete with a report on what the picture means.

Here is how it works, broken down into simple steps:

1. The Two-Phase Assembly Line

ARACRA splits the work into two main shifts, with a "quality control inspector" standing between them.

• Phase 1: The Clean-Up Crew (Raw Data to Counts)
  Imagine you have a bucket of muddy water (raw data from a chemical experiment). Phase 1 is the filtration system. It:

  • Fetches the water: It can pull data from a public library (the internet) or take it from your own bucket.
  • Filters the mud: It removes dirt, broken pieces, and bad data (like a coffee filter).
  • Sorts the pieces: It lines up the puzzle pieces in the right order.
  • The Inspector: Before moving to the next phase, a human expert gets to look at the filtered water. If the water is still too muddy, they can throw out specific samples. This ensures no bad data ruins the final result.

• Phase 2: The Detective Work (Finding the Patterns)
  Now that the data is clean, Phase 2 starts the real investigation:

  • Spotting the changes: It compares the "clean" cells to "dirty" cells to see which instructions changed.
  • The Dose-Response Curve: This is like testing how much poison it takes to make a plant wilt. ARACRA calculates exactly how much of a chemical is needed to start changing the cell's behavior.
  • The "Point of Departure" (tPoD): This is the most important number. It's the "tipping point." It tells regulators: "If you expose humans to more than X amount of this chemical, the body's instructions will start to go wrong."

2. Why is ARACRA Special?

Before ARACRA, scientists had to use a "Swiss Army Knife" approach. They would use one tool to clean data, a different tool to sort it, a third tool to do the math, and a fourth to draw the graphs. Every time they switched tools, they had to manually copy-paste data, which is like translating a book from English to French, then to Spanish, then back to English—you lose meaning and make mistakes along the way.

ARACRA is the "All-in-One Kitchen Robot."

• No Command Line Required: You don't need to be a coding wizard. You just use a simple website (like a dashboard in a car) to upload your data and click "Start."
• Human-in-the-Loop: It doesn't just run blindly. It pauses to let a human expert check the work, ensuring the computer doesn't make a silly mistake.
• Chemical Safety Focus: It is specifically built for chemical risk assessment. It doesn't just tell you what changed; it tells you how dangerous the change is at different levels of exposure.

3. The Real-World Test

The authors tested ARACRA on a dataset involving Bisphenol A (BPA) and 11 alternative chemicals (often found in plastics).

• The Result: ARACRA successfully processed hundreds of samples and identified which chemicals were the most toxic.
• The Match: It gave results that matched what other experts had found using much more complicated methods. For example, it confirmed that 2,4'-BPA was more active (and potentially more dangerous) than regular BPA, and that some other chemicals were essentially harmless.

The Bottom Line

ARACRA is a bridge. It connects the complex, messy world of raw biological data with the clear, actionable decisions needed to protect public health.

• For the Scientist: It saves hours of coding and prevents errors.
• For the Regulator: It provides a clear, standardized "stop sign" (the tPoD) to know when a chemical is unsafe.
• For You: It means that the chemicals in your daily life are being checked with a faster, more reliable, and more transparent system.

In short, ARACRA turns the chaotic noise of genetic data into a clear, loud alarm bell that tells us exactly when a chemical is crossing the line from "safe" to "risky."

Technical Summary

1. Problem Statement

Transcriptomic analysis is a critical tool for biomarker discovery and chemical risk assessment, particularly for deriving transcriptomic Points of Departure (tPoDs) and Benchmark Doses (BMD). However, the current landscape presents significant challenges:

• Fragmentation: Existing workflows require the sequential use of multiple independent tools (e.g., one for alignment, another for differential expression, and a third for dose-response modeling). This necessitates manual data format conversions, parameter re-specification, and increases the risk of inconsistency.
• Lack of End-to-End Solutions: No single existing tool spans the entire workflow from raw sequencing reads (FASTQ) to regulatory-quality pathway-level tPoDs.
• Specialized Data Gaps: General-purpose pipelines often fail to support targeted sequencing platforms like TempO-Seq, which require probe-specific alignment and aggregation strategies.
• Accessibility: Complex command-line interfaces and computational requirements hinder non-bioinformatician toxicologists and risk assessors from performing robust analyses.

2. Methodology: ARACRA Architecture

ARACRA (Automated RNA-seq Analysis for Chemical Risk Assessment) is a fully automated, reproducible pipeline designed to bridge these gaps. It is built on Nextflow DSL2 for workflow management, R for statistical analysis, and Streamlit for an interactive web interface.

The pipeline operates in two distinct phases separated by a "Human-in-the-Loop" (HITL) quality control gate:

Phase 1: Data Acquisition, Preprocessing, and Quantification

• Data Retrieval: Supports downloading raw data from the Sequence Read Archive (SRA) via accession numbers or uploading local FASTQ files.
• Pre-alignment QC: Uses fastp for adapter trimming and quality filtering (generating per-sample reports). Optional contamination screening is performed using FastQ Screen.
• Alignment:
  • Supports HISAT2 (low resource) and STAR (high resource) for standard RNA-seq.
  • TempO-Seq Support: Implements a unique dynamic indexing strategy where a custom, miniature FASTA is generated from user-provided probe manifests. STAR is configured to disable splicing and restrict mismatches for precise probe-level mapping.
• Quantification: Uses featureCounts (for alignment-based) or Salmon (for pseudo-alignment/quasi-mapping). For TempO-Seq, a custom Python script aggregates probe counts to target genes.
• Post-Alignment QC: Integrates RSeQC, Picard, and Qualimap. Samples are automatically flagged or removed based on thresholds (e.g., <70% reads uniquely mapped or >70% reads with Q<30).
• Exploratory Analysis: Performs VST normalization, batch correction (ComBat-seq), and PCA. Outliers are identified based on Euclidean distance in PC space, allowing users to visually inspect and decide on sample inclusion/exclusion.

Phase 2: Statistical Analysis and Dose-Response Modeling

• Differential Expression (DE): Uses DESeq2 for DE analysis, handling batch effects and treatment covariates.
• Dose-Response (DR) Modeling: Utilizes the DRomics R package.
  • Fits genes to linear, quadratic, or ANOVA models.
  • Selects the best-fitting curve (sigmoidal, exponential, etc.) using AICc.
  • Calculates Benchmark Dose (BMD) values using bootstrap confidence intervals.
• Quality Filtering: Applies rigorous filters to ensure reliability:
  • Max-dose filter: Removes genes requiring extrapolation beyond tested doses.
  • BMDU/BMDL ratio filter: Removes genes with wide confidence intervals (poor precision).
  • Fold-change pre-filter: Ensures biological relevance.
• tPoD Derivation:
  • Pathway-based: Maps BMD genes to GO, KEGG, and MSigDB collections. The tPoD is the median BMD of the most sensitive qualifying pathway.
  • Distribution-based: Calculates tPoDs based on ranked gene distributions (e.g., 25th ranked gene BMD, 5th/10th percentiles).

3. Key Contributions

• Unified Workflow: The first tool to integrate raw read processing, QC, DE, and dose-response modeling into a single, reproducible Nextflow pipeline.
• Human-in-the-Loop Review: Unlike fully automated "black box" systems, ARACRA includes an interactive QC gate where users can visually inspect data quality and manually curate outlier samples before proceeding to statistical analysis.
• TempO-Seq Compatibility: Specifically addresses the analytical needs of targeted TempO-Seq data, a growing standard in high-throughput toxicogenomics, which is often unsupported by general RNA-seq pipelines.
• Accessibility: Provides a Streamlit-based web interface that allows users to run complex analyses, tune parameters, and visualize results without command-line expertise.
• Regulatory Alignment: Adheres to regulatory standards (e.g., NTP 2018 guidelines, R-ODAF framework) for deriving tPoDs, making it suitable for chemical risk assessment.

4. Results and Validation

The pipeline was validated using a public TempO-Seq dataset (GSE271332) from Beal et al. (2024), analyzing Bisphenol A (BPA) and 11 alternatives in MCF-7 cells.

• Performance: Successfully processed 286 samples. For BPA, it identified 1,181 dose-responsive genes, fitted 892 models, and retained 282 high-quality BMDs after filtering.
• Concordance: The derived tPoD ranges and chemical potency rankings (e.g., 2,4'-BPA being more potent than BPA) were consistent with the original study and the R-ODAF pipeline.
• Biological Relevance: Identified known endocrine disruption genes (e.g., GREB1, PGR) associated with estrogen receptor pathways, confirming the biological validity of the results.
• Filtering Efficacy: The BMDU/BMDL ratio filter was the most effective at removing low-precision estimates, removing ~62.5% of initial BMDs for BPA, ensuring high-confidence downstream results.

5. Significance

ARACRA represents a significant advancement in computational toxicology and regulatory science:

• Standardization: It promotes harmonized data processing, metadata handling, and reporting, which is crucial for cross-study comparability in projects like the EU-PARC.
• Democratization: By abstracting computational complexity into a user-friendly interface, it enables toxicologists and risk assessors to perform high-level transcriptomic risk assessments without needing deep bioinformatics skills.
• Reproducibility: The use of Nextflow and containerization (Conda/Mamba) ensures that every step, parameter, and decision is recorded and reproducible, addressing a major pain point in omics research.
• Future-Proofing: The architecture is designed to evolve, with planned integration of Adverse Outcome Pathways (AOPs), multi-species support, and AI-assisted analysis via Model Context Protocol (MCP) servers.

Availability: ARACRA is freely available on GitHub under the MIT License, with a web-based demo and comprehensive user documentation.