From wiki
Jump to: navigation, search


The widely established tool for transcriptome assembly. It consists of three main components (which is how it got its name) although a fourth one, Jellyfish, is also included in the following list because it's important, thogh not written by the Trinity team:

  • Jelly fish (building a k-mer catalog from reads)
  • Inchworm (Linear contig construction from k-mers)
  • Chrysalis (Contig Clustering & de Bruijn Graph Construction)
  • Butterfly (also called Parafly, referred to as Phase 2, assembling clusters of reads)

In turn these 4 components can be grouped into two major phases:

  • Phase 1: Clustering of RNA-Seq Reads: this is made up of the first three components
  • Phase 2: Assembling Clusters of Reads: this is the Butterfly/Parafly stage

This list follows the sequence of operations.

Version 2.3.2, full help file

This is an edited version. With the module loaded, type

Trinity --show-full-usage-info

For the raw version.

Obligatory options

  • --seqType <string>, what type of reads the inputs are: ('fa' or 'fq')
  • --max_memory <string>, suggested max memory to use by Trinity where limiting can be enabled. (jellyfish, sorting, etc) in Gb of RAM, ie. '--max_memory 10G'
  • --left <string>  :left reads, one or more file names (separated by commas, no spaces)
  • --right <string>  :right reads, one or more file names (separated by commas, no spaces)
  • --single <string> (obviously neither obligatory nor necessary if paired reads are being input). Ssingle reads, one or more file names, comma-delimited (note, if single file contains pairs, can use flag: --run_as_paired )


  • --samples_file <string> tab-delimited text file indicating biological replicate relationships. For example:
cond_A    cond_A_rep1    A_rep1_left.fq    A_rep1_right.fq
cond_A    cond_A_rep2    A_rep2_left.fq    A_rep2_right.fq
cond_B    cond_B_rep1    B_rep1_left.fq    B_rep1_right.fq
cond_B    cond_B_rep2    B_rep2_left.fq    B_rep2_right.fq

(if single-end instead of paired-end, then leave the 4th column above empty)


  • --SS_lib_type <string> Strand-specific RNA-Seq read orientation. If paired: RF or FR. If single: F or R. For dUTP method = RF.
  • --CPU <int>, number of CPUs to use, default: 2
  • --min_contig_length <int>, minimum assembled contig length to report (default=200)
  • --long_reads <string>, fasta file containing error-corrected or circular consensus (CCS) pac bio reads (Note: experimental parameter, this functionality continues to be under development)
  • --genome_guided_bam <string>, genome guided mode, provide path to coordinate-sorted bam file (see genome-guided param section below)
  • --jaccard_clip, set if you have paired reads and you expect high gene density with UTR overlap (use FASTQ input file format for reads). Note that this is an expensive operation, so avoid using it unless necessary due to finding excessive fusion transcripts without it.)
  • --trimmomatic, run Trimmomatic to quality trim reads (see option --quality_trimming_params).
  • --no_normalize_reads, a major change in version 2.3.2, because normalization is now the default. Defaults to max. read coverage of 50. See option --normalize_max_read_cov' below.
  • --no_distributed_trinity_exec, do not run Trinity phase 2 (assembly of partitioned reads), and stop after generating command list.
  • --output <string>, name of directory for output (will be created if it doesn't already exist). Default is your current working directory
  • --full_cleanup, only retain the Trinity fasta file, rename as ${output_dir}.Trinity.fasta
  • --cite, show the Trinity literature citation
  • --verbose, provide additional job status info during the run.
  • --version, reports Trinity version (Trinity-v2.3.2) and exits.
  • --show_full_usage_info, show more options for running Trinity (expert usage).
  • --KMER_SIZE <int>, kmer length to use (default: 25) max=32
  • --prep, only prepare files (high I/O usage) and stop before kmer counting.
  • --no_cleanup, retain all intermediate input files.
  • --no_version_check, dont run a network check to determine if software updates are available.

Inchworm and K-mer counting-related options

  • --min_kmer_cov <int>, min count for K-mers to be assembled by Inchworm (default: 1)
  • --inchworm_cpu <int>, number of CPUs to use for Inchworm (only!), default is min(6, --CPU option)
  • --no_run_inchworm, stop after running jellyfish, before inchworm. (phase 1, read clustering only)

Chrysalis-related options

  • --max_reads_per_graph <int>, maximum number of reads to anchor within a single graph (default: 200000)
  • --min_glue <int>, min number of reads needed to glue two inchworm contigs together. (default: 2)
  • --no_bowtie, dont run bowtie to use pair info in chrysalis clustering.
  • --no_run_chrysalis, stop after running inchworm, before chrysalis. (phase 1, read clustering only)

Butterfly-related options

  • --bfly_opts <string>, additional parameters to pass through to butterfly see butterfly options:
java -jar Butterfly.jar

(note: only for expert or experimental use. Commonly used parameters are exposed through this Trinity menu here).

  • --group_pairs_distance <int>, maximum length expected between fragment pairs (default: 500). Reads outside this distance are treated as single-end.

Butterfly default reconstruction mode settings

Note: no CuffFly or PasaFly (quite a few errors in the help file .. this is Parafly really) custom settings are currently available)

  • --path_reinforcement_distance <int>, minimum overlap of reads with growing transcript path (default: PE: 75, SE: 25). Set to 1 for the most lenient path extension requirements.

Butterfly transcript reduction settings:

  • --no_path_merging, all final transcript candidates are output (including SNP variations, however, some SNPs may be unphased)

By default, alternative transcript candidates are merged (in reality, discarded) if they are found to be too similar, according to the following logic:

(identity=(numberOfMatches/shorterLen) > 95.0% or if we have <= 2 mismatches) and if we have internal gap lengths <= 10

with parameters as:

  • --min_per_id_same_path <int>, default: 98 min percent identity for two paths to be merged into single paths
  • --max_diffs_same_path <int>, default: 2. Max allowed differences encountered between path sequences to combine them
  • --max_internal_gap_same_path <int>, default: 10. Maximum number of internal consecutive gap characters allowed for paths to be merged into single paths.

If, in a comparison between two alternative transcripts, they are found too similar, the transcript with the greatest cumulative compatible read (pair-path) support is retained, and the other is discarded.

Butterfly, Java and parallel execution settings

  • --bflyHeapSpaceMax <string>, java max heap space setting. Default is 4G. Actually this is just the java -Xmx option. 4G causes java --Xmx4G.
  • --bflyHeapSpaceInit <string>, java initial heap space settings. Default: 1G. Actually the java -Xms option.
  • --bflyGCThreads <int>, threads for garbage collection (default: 2))
  • --bflyCPU <int>, CPUs to use (default will be normal number of CPUs; e.g., 2)
  • --bflyCalculateCPU, Calculate CPUs based on 80% of max_memory divided by maxbflyHeapSpaceMax
  • --bfly_jar <string>, /path/to/Butterfly.jar, otherwise default Trinity-installed version is used.

Quality Trimming Options

  • --quality_trimming_params <string> defaults to: "ILLUMINACLIP:/usr/local/Modules/modulefiles/tools/trinity/2.3.2/trinity-plugins/Trimmomatic/adapters/TruSeq3-PE.fa:2:30:10 SLIDINGWINDOW:4:5 LEADING:5 TRAILING:5 MINLEN:25"

In silico Read Normalization Options

  • --normalize_max_read_cov <int>, defaults to 50
  • --normalize_by_read_set, run normalization separate for each pair of fastq files, then one final normalization that combines the individual normalized reads. Consider using this if RAM limitations are a consideration.

Genome-guided de novo assembly

  • required: --genome_guided_max_intron <int>, maximum allowed intron length (also maximum fragment span on genome)
  • optional: --genome_guided_min_coverage <int>, minimum read coverage for identifying and expressed region of the genome. (default: 1)
  • optional: --genome_guided_min_reads_per_partition <int>, default min of 10 reads per partition

Trinity phase 2 (parallel assembly of read clusters) Options

  • --grid_exec <string>, your command-line utility for submitting jobs to the grid. This should be a command line tool that accepts a single parameter:
${your_submission_tool} /path/to/file/containing/commands.txt

and this submission tool should exit(0) upon successful completion.

  • --grid_node_CPU <int>, number of threads for each parallel process to leverage. (default: 1)
  • --grid_node_max_memory <string>, max memory targeted for each grid node. (default: 1G)

Note: The --grid_node_CPU and --grid_node_max_memory are applied as the --CPU and --max_memory parameters for the Trinity jobs run in Trinity Phase 2 (assembly of read clusters).

A typical Trinity command might be

Trinity --seqType fq --max_memory 50G --left reads_1.fq  --right reads_2.fq --CPU 6

and for Genome-guided Trinity:

Trinity --genome_guided_bam rnaseq_alignments.csorted.bam --max_memory 50G --genome_guided_max_intron 10000 --CPU 6
  • see:

for sample data and '' for example Trinity execution