Bismark alignment failure (0% mapping) caused by Bowtie2 2.5.5 regression with EM-seq paired-end parameters

Description of the bug

Description

Rerunning nf-core/methylseq 4.1.0 with -profile conda on identical data and parameters as a previously successful run (November 2024) resulted in ~0% alignment for all samples. The pipeline completed without errors.

Root cause

Conda silently upgraded Bowtie2 from v2.5.4 to v2.5.5 (published to bioconda ~February 2026) between runs. Since the bismark/align environment.yml only pins bismark=0.25.0 without an explicit Bowtie2 version constraint, the newer version was resolved automatically.

Bowtie2 2.5.5 appears to produce near-zero alignments when used with EM-seq paired-end parameters (--no-mixed --no-discordant --dovetail --maxins 1000). The same FASTQ files aligned at ~91% with Bowtie2 2.5.4 and 0% with 2.5.5.

Workarounds

Pin bowtie2=2.5.4 in modules/nf-core/bismark/align/environment.yml
Use -profile singularity instead of -profile conda

Suggested fix

Pin Bowtie2 to v2.5.4 in the bismark module environments until the regression in v2.5.5 is resolved upstream.

Environment

nf-core/methylseq: 4.1.0
Bismark: 0.25.0
Bowtie2: 2.5.5 (broken) / 2.5.4 (working)
Nextflow: 25.04.6
Profile: conda
Aligner: bismark
EM-seq: true

Command used and terminal output

Relevant files

No response

System information

No response

Start with modules/nf-core/bismark/align/environment.yml and compare its Bowtie2 dependency with the reported working and failing versions. Confirm the environment resolves the intended version, then verify that the EM-seq paired-end alignment no longer exhibits the reported near-zero mapping.