Reads or references
Build a reference graph from FASTA, or a cutoff-filtered read graph from FASTQ. Gzipped input is handled directly.
Cuttlefish
Fast, parallel, low-memory construction of compacted de Bruijn graphs — from reference sequences or sequencing reads, at collection scale.
Two lines of Cuttlefish
Section titled “Two lines of Cuttlefish”Cuttlefish has been through three generations. Which documentation you want depends on which one you are running.
Cuttlefish 3 — the Rust implementationThe current generation and canonical implementation. Uncolored and colored compacted graphs, external memory throughout, built for collections that do not fit in RAM.
Cuttlefish 1 & 2 — the C++ implementationThe earlier generations, widely deployed and installable from Bioconda. Cuttlefish 1 emits GFA and sequence tilings; Cuttlefish 2 handles both reads and references. Also covers the initial C++ implementation of Cuttlefish 3.
Not sure which applies to you? See which version should I use?.
What Cuttlefish does
Section titled “What Cuttlefish does”Cuttlefish constructs the compacted de Bruijn graph — the graph whose vertices are the maximal unitigs, the non-branching paths of the de Bruijn graph — from a set of reference sequences or sequencing reads. It is designed to stay fast and lightweight as the input grows to collection scale.
Colored graphs
Cuttlefish 3 emits positional colors: which input sources cover which stretch of each unitig, written alongside the unitig FASTA.
External memory
Partition buckets, blocked edge matrices, and collation buckets live on disk, so peak RAM tracks the worker set rather than the graph.
Parallel throughout
Partitioning, local contraction, discontinuity contraction, and final collation all run multi-threaded under a worker budget you set.
Citing Cuttlefish
Section titled “Citing Cuttlefish”If Cuttlefish contributed to your work, please cite the relevant paper — there is a separate one for Cuttlefish 1, 2, and 3.