Performance

Cis-GS was benchmarked on the Arachis hypogaea Tifrunner genome (2.6 Gb, 106,608 annotated genes, 2 kb promoters) on one machine: Intel Core i5-14450HX (16 logical cores), 18 GB RAM, WSL2 Ubuntu. Every measurement is a fresh process, three repetitions, warm file cache; peak memory is the operating system’s maximum resident set size. Reference tools: bedtools 2.31.1, seqkit 2.14.0, samtools 1.24, pyfaidx 0.9.0.4, MEME Suite 5.5.9.

Cis-GS performance on the Arachis hypogaea genome

Resources per step

Step

Tool

Wall time

Peak RAM

Note

Promoter extraction

Cis-GS 1.3.2.4

4.9 s

0.39 GB

includes GFF3 parse; 20.8 s on a cold first read

bedtools getfasta

1.8 s CPU

18 MB

plus 1.4 s BED, one-off samtools faidx

seqkit subseq

3.7 s CPU

0.92 GB

pyfaidx loop

0.9 s CPU

18 MB

Motif scan (2 motifs, incl. significance)

Cis-GS

15.4 s

0.64 GB

161 hits

seqkit locate -d (2 runs)

20.2 s

61 MB

161 hits (identical)

FIMO, p <= 1e-4

68.7 s

44 MB

25,774 PWM hits (different question)

Short motif stress test (GGATT)

Cis-GS

13.0 s

1.18 GB

442,811 hit rows

Co-expression pipeline (5,000 genes x 54 samples)

Cis-GS

10.1 s

1.67 GB

2.06 M edges, 15 modules

Notes

  • Extraction output is identical across Cis-GS, bedtools, seqkit and pyfaidx (same 106,608 sequences).

  • bedtools and pyfaidx remain faster at pure interval extraction; Cis-GS additionally parses the GFF3 and writes the promoter table used by later steps.

  • Cores: extraction and motif scanning run on one core. The co-expression step uses several cores inside NumPy / scikit-learn (about five on average). A multiprocessing prototype of the scan reached 4.9x at 16 workers; it is not part of the release.

  • Memory: every step stayed under 2 GB, so the workflow runs on an 8 GB laptop.

  • FIMO is listed for timing only: it scores position weight matrices and accepts mismatches, whereas Cis-GS matches IUPAC consensus patterns.