Consect

Basic Usage

consect.sh in=<file,file,file,...> out=<file>

Consect requires a minimum of 3 input files: the first file must contain the original uncorrected reads, followed by at least two files containing corrected reads from different error-correction tools. All files must have reads in the same order.

Parameters

Consect accepts standard BBTools parameters and Java memory options. Parameters are organized by their function in the consensus generation process.

Examples

consect.sh in=original.fq,tadpole_corrected.fq,bless_corrected.fq out=consensus.fq

consect.sh in=raw.fq,corrected1.fq,corrected2.fq,corrected3.fq out=consensus.fq cq=t

consect.sh in=original.fq,spades_ec.fq,karect.fq out=consensus.fq verbose=t

Algorithm Details

Conservative Consensus Strategy

Consect implements a conservative consensus algorithm designed specifically for substitution corrections. The algorithm processes reads position by position, applying corrections only when all error-correction tools agree on the same base change.

Position-by-Position Analysis

For each position in a read, the algorithm:

• Counts base votes: Examines the base called by each correction tool at that position
• Requires unanimity: Only applies corrections when all tools agree on the replacement base
• Handles disagreements: When tools disagree, the original base is preserved
• Manages variable lengths: Handles cases where some reads are shorter than others

Quality Score Management

When the cq=t parameter is enabled:

• Quality scores at corrected positions are set to the maximum quality from all contributing tools
• Original quality scores are preserved at uncorrected positions
• This provides confidence estimates for the consensus corrections

Performance Characteristics

• Memory usage: Scales linearly with read length and number of input files
• Processing speed: Limited by I/O since all files must be read synchronously
• Accuracy: High precision (few false positives) due to conservative approach, but lower recall (some true corrections may be missed)

Statistical Reporting

Consect provides comprehensive statistics including:

• Total errors corrected and disagreements found
• Number of reads with corrections vs. disagreements
• Classification of reads as fully corrected, partially corrected, uncorrected, or error-free
• Processing throughput (reads and bases per second)

Design Limitations

Important constraints of the consensus algorithm:

• Substitutions only: Designed for base substitution corrections, not insertions or deletions
• Read order dependency: All input files must contain reads in identical order
• Minimum input requirement: Requires at least 3 files (original + 2 corrected versions)
• Conservative bias: Favors precision over recall, potentially missing some valid corrections

Output Statistics

Consect provides detailed statistics about the consensus process:

Correction Metrics

• Errors Corrected: Total number of positions where unanimous corrections were applied
• Disagreements: Total number of positions where correction tools disagreed

Read-Level Statistics

• Reads With Corrections: Number of reads that received at least one correction
• Reads With Disagreements: Number of reads where tools disagreed on at least one position
• Reads Fully Corrected: Reads with corrections but no disagreements
• Reads Partly Corrected: Reads with both corrections and disagreements
• Reads Not Corrected: Reads with disagreements but no corrections applied
• Reads Error Free: Reads with no corrections or disagreements

Best Practices

Input File Preparation

• Ensure all input files contain reads in identical order
• Use the same file format (FASTQ recommended) for all inputs
• Verify that correction tools used similar parameters and approaches

Tool Selection

• Choose diverse error-correction algorithms for better consensus reliability
• Consider tools with different algorithmic approaches (k-mer based, overlap based, etc.)
• Avoid tools that perform aggressive trimming or filtering that could affect read order

Parameter Tuning

• Enable cq=t when quality scores are important for downstream analysis
• Use verbose=t for initial runs to understand correction patterns
• Monitor memory usage with appropriate -Xmx settings for large datasets

Support

For questions and support:

• Email: bbushnell@lbl.gov
• Documentation: bbmap.org