…ap segfault

Same vroom mmap segfault fix as make_kendall_pairs.R (see prior commit
for full root cause description).

Replace readGeneric() with fread(select=c("chr","start","end","PeakName"))
+ makeGRangesFromDataFrame(starts.in.df.are.0based=TRUE). Only the 4
columns actually used are read; TargetGene and PairName are skipped.

Replace library(genomation) with library(data.table) for fread().

ATAC matrix rowname fix:
  Signac::FeatureMatrix derives rownames via GRangesToString, which
  emits 1-based GRanges starts. Under the old (buggy) code, the
  GRanges 1-based start numerically equalled the 0-based BED start,
  so rownames accidentally matched the 0-based PeakName strings in
  Pairs.Kendall.tsv.gz. With the corrected coordinate loading
  (starts.in.df.are.0based=TRUE), GRanges starts are now bed_start+1,
  so default Signac rownames would be chr-(a+1)-b while PeakName is
  chr-a-b.

  After FeatureMatrix, rownames are explicitly set to
  paste(seqnames, start-1L, end, sep="-") to restore 0-based BED
  naming. FeatureMatrix preserves input feature order, so this is a
  safe positional rename. compute_kendall.R matches
  mcols(bed.E2G)[,"PeakName"] %in% rownames(data.ATAC), so the two
  must agree on coordinate convention.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

…gfault

Same vroom mmap segfault fix as prior commits (see make_kendall_pairs
commit for full root cause description).

Replace readGeneric(kendall_pairs_path, keep.all.metadata=T, header=T)
with makeGRangesFromDataFrame(fread(kendall_pairs_path),
keep.extra.columns=TRUE, starts.in.df.are.0based=TRUE). All columns
from kendall_pairs are used in this script, so no select= narrowing.

Remove library(genomation) from imports.

Output coordinate preservation:
  df.pairs.E2G$start is decremented by 1 before fwrite, restoring
  0-based BED format in Pairs.Kendall.tsv.gz. This file is consumed
  by compute_arc_e2g.R, which expects 0-based BED coordinates.
  PeakName is carried through from the input file's metadata column
  (not reconstructed), so it remains in its original 0-based format.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

…gfault

This is the script that originally triggered the investigation: 1 of 4
scE2G runs segfaulted in vroom_() C++ code during the arc_e2g rule
(SIGSEGV with SEGV_ACCERR, "invalid permissions"). Root cause: vroom's
mmap-based lazy reading on Lustre, where lock revocation invalidates
mmap'd pages below R's memory management layer.

Replace readGeneric(abc_predictions_path) with
makeGRangesFromDataFrame(fread(abc_predictions_path),
keep.extra.columns=TRUE, starts.in.df.are.0based=TRUE). All ABC columns
are used (AddMax, IntegrateABC), so no select= narrowing.

Replace GRanges(fread(kendall_predictions_path)) with
makeGRangesFromDataFrame(fread(kendall_predictions_path),
keep.extra.columns=TRUE, starts.in.df.are.0based=TRUE). The old
GRanges(fread()) call also misinterpreted 0-based BED starts.

Remove library(genomation) from imports.

Coordinate correction:
  The old code loaded BED [a, b) as GRanges [a, b] (treating the
  0-based start as 1-based). This elongated every region by 1bp,
  causing findOverlaps in AddMax to produce spurious hits where two
  regions share a boundary but don't overlap in BED semantics.
  Quantified on igvf10/thp1_2: 1,502 of 10,303,415 ABC rows (0.015%)
  were affected; max ARC.E2G.Score ratio old/new = 2.05; max absolute
  score change = 0.0045.

Output coordinate preservation:
  df.output$start is decremented by 1 before fwrite, restoring 0-based
  BED format in EnhancerPredictionsAllPutative_ARC.tsv.gz.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>