PG-DIA

PG-DIA is a Nextflow DSL2 workflow for building customized protein databases from RNA-seq and searching matched DIA-MS data against those databases. This implementation combines RNA-seq variant calling, transcript assembly, novel isoform ORF prediction, protein database assembly, and novel peptides reporting in one pipeline.

What the pipeline does

For each sample, the workflow:

1. Accepts RNA-seq input as FASTQ, BAM, or CRAM together with one DIA raw file path.
2. Runs the RNA variant branch to produce BAMs and annotated VCFs.
3. Converts annotated variants into variant peptide FASTA entries with pypgatk, then annotates amino acid changes.
4. Runs StringTie on the RNA-seq alignment output.
5. Runs gffcompare, filters novel transcript models, extracts transcript FASTA with gffread, and predicts ORFs with TransDecoder.
6. Combines the reference proteome, variant peptides, and novel isoform peptides into per-sample protein databases.
7. Runs DIA-NN against the per-sample combined FASTA.
8. Post-processes the DIA-NN parquet report into reference and novel peptide/protein matrices.

This is a sample-matched workflow: each RNA-seq sample is paired to one DIA raw input and yields its own customized database and DIA-NN output.

A small simulated DIA-MS fixture is included for local file-ingestion and DIA-NN integration testing.

Inputs

Samplesheet

The full PG-DIA workflow expects one row per sample with:

• one RNA-seq input mode: fastq_1/fastq_2, or bam/bai, or cram/crai
• one DIA input path in dia_raw

An example samplesheet for this branch looks like this:

sample,fastq_1,fastq_2,bam,bai,cram,crai,dia_raw,strandedness
SAMPLE_A,/data/rna/SAMPLE_A_R1.fastq.gz,/data/rna/SAMPLE_A_R2.fastq.gz,,,,,/data/dia/SAMPLE_A.d,unstranded
SAMPLE_B,,,/data/rna/SAMPLE_B.markdup.bam,/data/rna/SAMPLE_B.markdup.bam.bai,,,/data/dia/SAMPLE_B.RAW,unstranded
SAMPLE_C,,,,,/data/rna/SAMPLE_C.cram,/data/rna/SAMPLE_C.cram.crai,/data/dia/SAMPLE_C.d,unstranded

Column notes:

Column	Required	Description
sample	Yes	Sample identifier. Multiple FASTQ lanes for the same sample may reuse the same name.
fastq_1, fastq_2	Conditional	RNA-seq FASTQs. Use these if starting from raw reads.
bam, bai	Conditional	Coordinate-sorted BAM and index if alignment is already done.
cram, crai	Conditional	CRAM and CRAI if starting from CRAM instead of BAM.
dia_raw	Yes	DIA-MS raw path for the sample, for example a timsTOF .d directory or vendor raw file.
strandedness	Yes	Passed to StringTie. Supported values are forward, reverse, and unstranded.

Required reference and workflow parameters

At minimum, plan to provide:

• --input
• --outdir
• --protein_reference_db
• either --genome with a configured reference bundle, or explicit --fasta and --gtf
• --read_length when aligning from FASTQ

For the RNA variant branch:

• provide --dbsnp and/or --known_indels if you want base recalibration
• otherwise set --skip_baserecalibration

Quick start

For a repository-local test using simulated RNA-seq, variant, reference, and DIA-MS inputs, follow the runnable example tutorial:

nextflow run main.nf -profile test,docker --outdir results/test

The tutorial includes DIA-NN image and executable-path configuration, a configuration preview, expected outputs, and fixture-regeneration commands.

For user data, run from the repository root:

nextflow run bzhanglab/pgdia \
  -profile docker \
  --input samplesheet.csv \
  --outdir results \
  --genome GRCh38 \
  --read_length 151 \
  --protein_reference_db /path/to/reference_proteome.fa \
  --skip_baserecalibration

If you have known sites for GATK RNA recalibration, use them instead of skipping BQSR:

nextflow run bzhanglab/pgdia \
  -profile docker \
  --input samplesheet.csv \
  --outdir results \
  --genome GRCh38 \
  --read_length 151 \
  --protein_reference_db /path/to/reference_proteome.fa \
  --dbsnp /path/to/dbsnp.vcf.gz \
  --known_indels /path/to/known_indels.vcf.gz

Useful runtime options:

• -profile docker, -profile singularity, -profile conda, or -profile mamba
• -resume to continue from a previous run
• --diann_image to point to a DIA-NN container image name or tarball; download the DIA-NN 2.0.2 image
• --diann_bin if the DIA-NN executable path inside the image differs from the default
• --diann_cpus to control DIA-NN threads

Paired-entrapment assessment

Add --paired_entrapment to run a separate per-sample assessment alongside the standard search. PG-DIA digests the reference and novel FASTAs into 7–30 residue peptides, collapses I/L-equivalent novel identities, and generates one random composition-preserving shuffle per retained novel target with its final residue fixed. Shuffles matching anywhere in either input FASTA are rejected. Proven unshufflable targets are excluded with reasons. The seed defaults to 20260817; --entrapment_missed_cleavages defaults to 1 and accepts 0, 1, or 2.

The separate DIA-NN search disables cleavage and methionine excision, retains search decoys, and treats entrapments as search targets. The paired FDP assessment uses Global.Q.Value as its q-value source. By default, complete Global.Q.Value score ties enter cumulative D/T together and reverse cumulative minima give empirical precursor q-values. --entrapment_cutoff defaults to 0.001. Use --entrapment_recalculate_q false to apply the cutoff directly to the supplied Global.Q.Value values.

The assessment collapses charges, modifications, and runs to unmodified peptide sequences. It reports (N_E + N_E_only + 2*N_E_better_both)/(N_T + N_E) for the combined novel-target/entrapment list. Tied accepted partners cause an error because their ordering is unresolved. This FDP estimate does not establish novel-target-only FDR at the nominal precursor cutoff. Outputs, including the search report, pair mapping, input hashes, exclusions, summary, and method metadata, are published under paired_entrapment/<sample>/.

Reference configuration

The bundled conf/igenomes.config defines specific --genome GRCh38 entry pointing to:

• genome FASTA
• annotation GTF
• STAR index
• VEP cache metadata

If you do not want to use that bundle, provide explicit reference files with parameters such as:

• --fasta
• --fasta_fai
• --dict
• --gtf
• --star_index
• --dbsnp
• --known_indels

--read_length matters for STAR index generation and alignment. Set it to the actual read length of the RNA-seq data, for example 151 for 2x151 bp libraries.

Main outputs

Published results are written under --outdir and typically include:

Path	Contents
reports/	MultiQC output and annotation summary reports
pipeline_info/	Nextflow execution metadata, params, and software versions
annotation/<sample>/	Annotated and decompressed VCF outputs from the RNA variant branch
stringtie/	Per-sample StringTie GTF assemblies
isoform_db/<sample>/	Predicted peptide FASTA from novel isoforms
protein_db/<sample>/	<sample>_combined_protein_db.fa and <sample>_novel_protein_db.fa
diann_output/<sample>/	DIA-NN parquet report, matrices parquet, per-run transition XIC parquet files, and postprocessed novel/reference TSV matrices

DIA-NN is run with --xic, which publishes transition-level chromatograms as diann_output/<sample>/<sample>_report_xic/*.xic.parquet. Use bin/plot_diann_xic.py to extract one precursor from these files, export its transition-level values, and generate per-run and combined XIC plots. See the output documentation for the input columns and a reproducible command.

Credits

PG-DIA in this repository was written and adapted by Wenrong Chen in the Zhang Lab, building on nf-core components.

Citation

See CITATIONS.md for software and workflow citation details.