Adapter trimming and the adapter screen¶

An Oxford Nanopore (ONT) read starts with the sequencing adapter, and in a barcoded library the barcode follows it. Left in the reads, both can end up at the ends of the assembled contigs, most often just outside the telomere. FunFlux therefore removes them from ONT reads before assembly, with fastplong, and searches every assembly for adapter sequence afterwards, in every mode.

Dorado removes adapters and barcodes when it is run by hand (dorado basecaller, kit14 kits), but reads basecalled live in MinKNOW, or with dorado duplex, usually keep them. A FASTQ file does not say which, so FunFlux looks at the reads themselves. PacBio HiFi reads are not trimmed: the PacBio software that makes them moves reads with adapter sequence to a separate file of failed reads.

---
config:
  themeVariables:
    fontSize: 14px
  flowchart:
    nodeSpacing: 35
    rankSpacing: 35
---
flowchart TD
    R([raw ONT reads]) --> S[find the adapter and the block after it<br/>find_ont_adapter<br/>all reads]
    S --> T[fastplong removes exactly that sequence<br/>trim_ont_adapters]
    T --> F[Filtlong]
    F --> C[count the adapters left<br/>check_ont_trimming]
    F --> H[hifiasm or Flye]
    H --> A([draft assembly])
    A --> B[search the contigs for adapters and barcodes<br/>blast_adapters, screen_adapters]
    S -.->|the trimmed sequence| B
    B --> Q([adapter column of the contig summary<br/>Adapters table of the MultiQC report])

Terms used on this page


adapter	the short synthetic DNA that brings a molecule to the nanopore; ONT's ligation, native and rapid adapters share the top strand `CCTGTACTTCGTTCAGTTACGTATTGCT`
barcode block	in a barcoded library, the sequence after the adapter: for native barcoding `AAGGTTAA`, the 24 bp barcode and `CAGCACCT`
trimmed sequence	what fastplong removes from the reads of one sample: the adapter, the adapter and its barcode block, or a sequence you set
chimeric read	one read that holds two molecules joined end to end, with an adapter between them
end hit	adapter sequence within 100 bp of a contig end
telomere	the tandem repeat (TTAGGG in most filamentous fungi) at the end of every chromosome
rule	one step of the workflow as Snakemake defines it, such as `trim_ont_adapters`; the console and the log files in `logs/` use its name

Settings¶

The key goes in the block named after the mode, parameters.longread or parameters.hybrid:

`ont_adapter`	Effect
`detect` (the default; also empty)	FunFlux works the sequence out from the reads, or leaves reads that are already trimmed as they are
a sequence of 20 to 200 bases of A, C, G and T	that sequence is trimmed, as written
`off` (also `false`, `no`, `none`)	no trimming: Filtlong reads the raw reads, as in FunFlux v2.0.0

A sequence that is a short unit repeated end to end, such as a telomere repeat, is refused when the workflow starts. A sequence shorter than 50 bp gets a NOTE, because fastplong then trims it only at the read ends (see below). The contig adapter screen has no thresholds to set.

How the sequence is found¶

Rule find_ont_adapter reads every read of the sample once; on the 5.8 Gb Botrytis cinerea B05.10 file this took about two minutes.

The adapter. The first 150 bp of every read are searched for the 20 bp core of the adapter top strand, TTCGTTCAGTTACGTATTGC. The core must match exactly, so reads whose adapter carries a basecalling error there are not counted: the share of reads with the adapter is a lower bound.
The block after it. In a barcoded library most reads continue after the adapter with the same barcode block, while in a library without barcodes they continue into the genome, which differs from read to read. FunFlux takes the longest block, up to 120 bp, that at least 60 % of the reads with the adapter share. In B05.10 the 40 bp native barcode block was shared by 87 % of them; the most common 44 bp sequence, by 1 %.
ONT's published barcodes. The sequence after the adapter is compared with the 96 native and 96 rapid barcodes that ONT publishes in its Chemistry Technical Document; a read counts for a barcode when at most 3 of the 24 barcode bases differ. In B05.10, 92.9 % of the reads with the adapter carry native barcode NB24, and no other barcode reaches 1 %. The list names the barcode and sorts the reads of a file that holds several barcodes; the sequence trimmed for each barcode is still the block found in its own reads. ONT publishes the rapid barcodes only in part, so the reads are the safer source, and a kit or barcode the list lacks is trimmed all the same.
The decision.
- When fewer than 1 % of reads start with the adapter, and fewer than 1 % start with a barcode block, the reads count as already trimmed and pass through unchanged.
- When published barcodes are found, each in at least 1 % of the reads with the adapter, the adapter followed by each barcode's block is trimmed. Two or more barcodes above 10 % get a NOTE, because such a file may hold several samples; a barcode below 10 % is usually a few reads of a neighbouring sample assigned to this one, and it is trimmed without a note.
- Otherwise the trimmed sequence is the adapter followed by the block, if there is one, or the adapter alone (a library without barcodes).
- When the adapter is missing but at least 1 % of reads start with a native or rapid barcode block (the adapter was removed earlier, the barcode left), each such block is trimmed.
A block that is a short tandem repeat is never added. A block with barcode flanks around a barcode that ONT does not publish is trimmed as found, with a NOTE. When the reads continue with two or more different blocks that are not published barcodes, only the part they share is trimmed, with a NOTE: those sequences then stay in the reads, and the adapter screen reports them. Write the whole sequence into ont_adapter to trim it.

The file 01.reads/{sample}/long/adapters/{sample}_adapter_scan.tsv (Output files) counts the reads with the adapter in every block of 10,000 reads. A batch of already-trimmed reads in an otherwise untrimmed file shows up there: in the public B05.10 file, reads 1 to about 95,000 carry almost no adapter, and most of the reads after them do.

What fastplong does with it¶

Rule trim_ont_adapters runs fastplong 0.7.1 with the most common sequence (-s), and with a FASTA file of every sequence to trim, in both orientations (-a). fastplong searches the first 200 bp of every read for the -s sequence and the last 200 bp for its reverse complement, which is how the barcode and adapter sit at the end of a native-barcoded read; the -a sequences are searched at both ends as written, which is how every barcode of a file that holds several is trimmed. fastplong allows a quarter of the bases to differ and removes the sequence together with everything outside it. FunFlux adds:

Option	Why
`--trimming_extension 0`	fastplong's default cuts 10 more bases past the sequence, which after a whole adapter and barcode are genome: on B05.10 it shortened every read that starts in a telomere
`-Q -L`	fastplong's quality and length filters stay off: Filtlong, the next step, is the read filter of the workflow
`--disable_chimera_splitting`, when a sequence to trim is shorter than 50 bp	fastplong also cuts a read where it finds the sequence in the middle (a chimeric read). With the 28 bp adapter alone, that split 1,041 B05.10 reads, and 1,001 of them were not chimeric; with the 68 bp adapter and barcode it split 72, and the 57 of them that could be checked were all genuine joins

Rule check_ont_trimming then counts the reads that still start with the adapter, among the reads kept after fastplong and Filtlong (the reads the assembler gets). Above 1 % it prints a NOTE; the run goes on.

fastplong's own guess of the adapter is never used. fastplong guesses from the first reads of the file only, and takes the most common sequence at the read ends. In fungal reads that can be the telomere repeat: on B05.10 it chose CCCTAA… and TTAGGG…, trimmed the telomeres from the reads, and the assembly kept a telomere at 2 contig ends instead of 34.

The contig adapter screen¶

Rule blast_adapters searches the draft contigs of every sample, in every mode, with BLAST (blastn -task blastn-short) for:

Sequences	Source
ONT adapters, the 96 native barcodes and the 96 rapid barcodes with their flanks	ONT Chemistry Technical Document (shipped with FunFlux)
PacBio and MGI adapters	NCBI UniVec (shipped with FunFlux)
Illumina adapters and primers	NCBI UniVec build 10.1, downloaded into the run, because Illumina's sequences are copyrighted
the sequence trimmed from the sample's reads	rule `find_ont_adapter` (trimmed ONT samples)

Rule screen_adapters keeps a hit when at least 90 % of its aligned bases are identical and it is at least 20 bp long within 100 bp of a contig end, or at least 28 bp long, a whole adapter, inside a contig. Overlapping hits form one locus. The end rule does not ask how much of an adapter record is covered, because adapter is often cut short at the very end of a contig, and a barcode whose adapter was trimmed must still count.

The result goes into the adapter column of the contig summary (none, at_end, internal or at_end+internal), into 02.assembly/{sample}/eval/adapter_screen.tsv with one row per locus, and into the Adapters table of the MultiQC report. The screen changes no contig and no verdict.

Before a GenBank submission

The lists cover the common adapters only. NCBI screens every genome submitted to GenBank with its own tool, FCS-adaptor, which is distributed as a container; run it on the genome before you submit.

What it did on Botrytis cinerea B05.10¶

The public B05.10 ONT reads (SRR40741277) are native-barcoded, and about 70 % of them were not trimmed at basecalling. FunFlux longread mode assembled them with hifiasm, as delivered and after trimming:

Reads given to hifiasm	Chromosomes in one contig	Contig ends with a telomere	Contigs with a telomere at both ends	Median distance from the telomere to the contig end	Contig ends with adapter or barcode
as delivered	18 of 18	34	15	78 bp	28
adapter only trimmed (28 bp)	18 of 18	35	17	72 bp	27
adapter and barcode trimmed (68 bp, the sequence FunFlux finds)	18 of 18	34	15	0 bp	0

Trimming the adapter and barcode removed them from every contig end and brought most telomeres to the very end of their contig. The telomere counts changed by no more than two. The last row is the FunFlux 2.1.0 validation run (Validation on public genomes). The first two are FunFlux 2.0.0 runs, on the reads as delivered and on reads trimmed beforehand with fastplong. Trimming also changed the mitochondrion: from the trimmed reads, hifiasm assembled it as a longer contig of more than one copy instead of the circle of 82,351 bp (Other results).

With Flye, the reads as delivered left adapter or barcode at 17 of 54 contig ends, and the trimmed reads at none. The public reference genome of B05.10 (ASM14353v4) gives no hit in the screen.

These numbers come from one strain, one native-barcoding kit and one barcode. Four public runs of the Rapid Barcoding Kit V14, one of them a yeast (Diutina catenulata, ERR17348184), had their adapters trimmed at basecalling: the scan left three unchanged, and with the yeast reads FunFlux produced the same assembly with trimming on and off. In the fourth (Bacillus cereus, ERR16717241) it found rapid barcode RB10 left at the start of 1.4 % of the reads and trimmed it. Untrimmed reads of a rapid or ligation kit have not been tested yet.