CycSim simulates long reads whose errors depend on the surrounding sequence. First cycsim train reads real reads aligned to a reference and learns, for each k-mer (11 bases by default), how often errors occur and of what type. It also learns read lengths. Then cycsim sim samples reads from a genome using that model. You give a trained model and a reference. It returns aligned reads in BAM format.
Key assumptions: error behavior is the same for the same k-mer context, and the read-length profile matches the training data. A model made from one chemistry does not carry over to another.
When to use it
Simulate reads for a chemistry that has a trained model: the README lists CycloneSEQ S1, CycloneSEQ HD118, PacBio HiFi and ONT Q20.
Train a model on your own run (HiFi or nanopore) to get errors that follow its context bias.
Compare against context-free simulators.
Alternatives on biosimhub: PBSIM3 uses quality-score HMMs for ONT and PacBio. SimLoRD models PacBio pass counts. PBSIM2 is the older HMM tool.
Install it yourself
CycSim 1.0.4 is on bioconda. It is built for linux/amd64 only. A dry run under emulation was killed, so the check is the recipe's image: it holds cycsim-1.0.4-h4eafe19_0 installed from the same channels.
-r takes hifi or nanopore. sim defaults from the help text: 3 threads, depth 30, minimum aligned length 1000, maximum 500,000, noise 0.000001, seed 0 (random). -D sets total bases instead of depth.
Run the tested example
This recipe builds CycSim in a container and runs the example below. It needs Docker and nothing else. It last ran on 2026-10-12 in 21.7 s. CycSim has no ARM build, so on Apple silicon Docker runs the x86 image under emulation, which is slower.
Download cycsim-recipe.tar.gz (Dockerfile, example model.mod, reference.fa, run.sh).
Inputs are read from in/ and results land in out/. Change the values after each option to run your own case; the parameter table lists them.
Expected output
File
Size
cycsim_log.txt
0 B
reads.bam
93.0 KB
Parameters
Option
Meaning
Example value
Range
-D
Total bases to simulate
100000
1000 to 200000000
-m
Minimum aligned read length
1000
100 to 100000
-M
Maximum aligned read length
500000
1000 to 1000000
-n
Proportion of random error bases
0.000001
0 to 0.1
-s
Seed
42
1 to 2147483647
Ranges are the limits used for the example recipe; the program itself may accept more.
Reading the output
The example uses a shipped model (model.mod, 154,172 bytes) and a 20,000 bp reference. -D 100000 asks for about 100,000 bases. The output reads.bam holds the simulated reads with their alignments; convert it with samtools fastq reads.bam if your pipeline wants FASTQ.
Decoded, the file has an unsorted header with one reference (synthetic_ref, 20,000) and 23 reads. They total 102,461 bases, with lengths from 2,516 to 5,955 (mean 4,455). Read names carry the origin, for example 000_1_synthetic_ref_6103_8957_0_2847_0_0_0. Alignment columns: 100,471 matches or mismatches, 1,990 insertions and 1,802 deletions, an indel rate of 3.6%. cycsim_log.txt is empty.
Your own runs
All runs use seed 42 and one thread.
Same seed, same reads. A second run gives an identical BAM after decompression. The help warns that read order can vary with several threads.
More bases.-D 400000. Result: 99 reads, 400,074 bases, indel rate unchanged at 3.6%.
Context-free errors. Add -g. Result: 25 reads, indel rate 3.7%. On this small example both modes give similar indel totals (3.7% and 3.6%).
Two notes for your own runs. The model file ends in .mod here and .cy in the README. The example loads the .mod file, so the extension does not matter to the tool. The log is empty because the tool prints nothing on success.
Pitfalls
Output is BAM. Name it .bam in your own runs so tools open it correctly.
Model and chemistry must match. Train on reads from the same platform you want to simulate.
Maximum length is a cap. The example sets -M 500000 but the longest read is 5,955 bp. The reference is 20 kb and the model sets the length profile.
Seed 0 means random. The default seed gives different reads each run.
Memory. Training loads up to 1 Gb of bases at once (-b). The recipe's container allows 3 GB.