This dataset accompanies the manuscript "Genome-wide mapping of genetic suppression events in Streptococcus pneumoniae" (Ng et al.). It contains the R code and the per-query statistical output used to identify genetic-suppression hits from Sup-seq (suppressor sequencing) data.

Sup-seq is a randomly-barcoded transposon-sequencing (RB-TnSeq) approach that detects suppressors of essential genes genome-wide. An essential query gene is deleted in a pre-mapped, barcoded Himar1 transposon (Tn) library; surviving colonies are enriched by backcrossing; and barcode abundances in the query pool are compared with an unselected control pool. Transposon insertions can suppress essentiality through loss-of-function, read-through overexpression, or antisense downregulation, so insertions on the two strands (orientations) are analyzed separately.

The files here implement the hit-calling statistics for the revised manuscript: a moderated t-like test statistic applied per region × strand, adapted from the original RB-TnSeq method (Wetmore et al. 2015, mBio, PMID:25968644).

Contents

Each tests.<query>.tsv file contains ~5,500 rows (roughly 3,100 unique loci × 2 strands, minus region × strands that did not pass the coverage filter).

Note on the eighth query. The manuscript reports statistics across 8 queries; the eighth is the control-vs-control comparison (control1 vs control2), used only to estimate the false-positive rate, and is not included as a tests.*.tsv file here.

Data file column definitions (tests.<query>.tsv)

Tab-separated, one row per region × strand.

Calling significant hits

A region × strand was called a significant suppressor hit when:

t >= 5   AND   fitnorm >= 2   AND   max >= 6

Applying this filter to tests.ackA.tsv, for example, yields 50 region × strand hits. To reproduce the manuscript's cluster-level counts, adjacent significant loci on the same strand (within 3 kb) are merged into a single cluster, because polar transposon effects frequently spread across neighboring loci that act on the same causative gene.

Methods summary

Per-strain fitness is the log2 ratio of query-pool reads to control-pool reads (summed across the two control libraries), with a small pseudocount added to stabilize counts of zero, normalized so that the median strain has fitness 0, and further normalized for position along the chromosome (position effects were moderate; maximum estimated effect on log2 ratios ≈ 0.50).

The t-like statistic is the coverage-weighted mean fitness divided by an estimated standard error, where the variance is a moderated mixture of (i) the observed consistency of fitness values within the region and (ii) the Poisson (count-based) expectation, borrowing information across regions to stabilize estimates for regions with few insertions. To estimate within-region consistency, barcodes at each region × strand are sorted by insertion position and assigned alternately to two groups (1st barcode → group 1, 2nd → group 2, …), rather than splitting the region into 5′ and 3′ halves as in the original RB-TnSeq method. This alternating scheme is used because many strong Sup-seq effects act on a downstream gene, so fitness varies systematically with insertion position within a region.

Inclusion thresholds: only strains with ≥ 4 total reads (query + controls) are analyzed, and only region × strands with ≥ 10 total control reads. High fitness is defined as per-strain fitness ≥ 6 with ≥ 10 reads.

False discovery rate

Applying the significance filter to the two control-vs-control comparisons yielded no hits. As a second FDR estimate, the experimental data were shuffled (control and experimental counts for each strain preserved, but genomic location randomized); across 10 shuffles this produced an average of ~20 hits per genome-equivalent, giving an estimated FDR below 5 %. Restricting to hits with ≥ 2 high-fitness strains (nHi >= 2) lowers the estimated FDR to ~2.5 %. See the manuscript Methods for the exact counts reported.

Software requirements

• R (developed and tested with base R; no version-specific features required).
• R package parallel (base R distribution) — FEBA.R uses mclapply(). The number of CPUs is controlled by the MC_CORES environment variable (default 2).
• No other external packages are required.

Memory note: FEBA.R is designed for large genome-scale datasets and can be memory-intensive on very large strain × experiment matrices (the header notes ~10 GB for ~200K strains × ~500 experiments). The per-query Sup-seq analyses in this dataset are much smaller.

Usage

FEBA.R and supseq.R are function libraries — source them, then call the Sup-seq routines with your count matrices:

source("FEBA.R")
source("supseq.R")

# meta: data frame with columns scaffold, strand, pos, locusId (one row per barcoded strain)
# counts: vector of query-pool read counts per strain
# controlCounts: vector of control-pool read counts per strain (sum of the two controls)

# Per-strand statistics for one query, both strands combined:
results <- SupSeqCombined(meta, counts, controlCounts)

# The output columns match the tests.<query>.tsv files: locusId, strand, n, nHi, max, t, fitnorm

show(cond, locusId=...) plots the per-strain log2 ratios in a ±3 kb window around a locus, colored and marked by insertion strand, with a gene map overlaid — useful for inspecting the insertion profile at a candidate hit.

The tests.<query>.tsv files in this dataset are the saved output of running SupSeqCombined() on each of the seven essential query genes; they can be used directly without re-running the analysis.

Related data

The raw barcode sequencing reads, the RB-TnSeq transposon library, and the RNA-seq and lipidomics data associated with the manuscript are deposited separately — see the Data Availability statement of the published article for repository accessions.

Citation

If you use these scripts or data, please cite the associated manuscript:

Ng, C. J. W., Zik, J. J., Li, Y., Chen, J.-H., Guan, X., Arkin, A. P., Deutschbauer, A. M., & Sham, L.-T. Genome-wide mapping of genetic suppression events in Streptococcus pneumoniae. [Journal, year, DOI — to be added upon publication].

The underlying RB-TnSeq fitness statistics implemented in FEBA.R are described in:

Wetmore, K. M., Price, M. N., Waters, R. J., et al. (2015). Rapid quantification of mutant fitness in diverse bacteria by sequencing randomly bar-coded transposons. mBio 6(3):e00306-15. PMID:25968644.

Contact

[Corresponding author name and email — e.g., Lok-To Sham, lsham@nus.edu.sg]