Short-Read, Long-Read or Hybrid: Choosing a Bacterial Genome Assembly Strategy

Why short-read assemblies are fragmented

An assembler rebuilds a genome by overlapping reads. When a sequence is repeated in several places, such as ribosomal RNA operons, insertion sequences or transposase genes, a read shorter than the repeat can't show which copy it came from. The assembler stops at each repeat, and the genome ends up in many pieces (contigs). Illumina reads are accurate but short, so they give accurate but fragmented assemblies. Long reads can produce complete assemblies but have historically been more error-prone ^[1].

Fragmentation matters most for the genes you often care about. Resistance genes, virulence genes and plasmids are frequently next to repeats and mobile elements, so they tend to land at the ends of contigs or get split between them.

The three strategies compared

Assembly tools for each approach

SPAdes is a standard choice for short-read bacterial assembly ^[2]. Unicycler was built for hybrid data: it creates an assembly graph from short reads with SPAdes, then uses long reads to resolve the graph into complete replicons. In tests it produced larger contigs with fewer misassemblies than other hybrid assemblers, even when long-read depth and accuracy were low ^[1].

For long reads alone, Flye builds a repeat graph that represents repeated regions explicitly. It was faster than competing assemblers and gave better or comparable results ^[3]. Because every long-read assembler makes some errors, Trycycler combines several independent assemblies of the same genome into a consensus. In benchmarks this produced fewer errors than any single tool, and adding polishing gave the most accurate genomes in the study ^[4]. Trycycler needs some manual work, so it suits projects where a finished reference genome is the goal.

Long-read accuracy has improved quickly. With Oxford Nanopore R10.4 chemistry, Sereika and colleagues generated near-finished microbial genomes from isolates and metagenomes without short-read or reference polishing ^[5]. For many projects a long-read-only genome is now a realistic option.

Which strategy fits your question?

• Identifying species and screening many isolates for resistance or virulence genes: short reads are usually enough and cost the least per sample.
• Finding out whether a resistance gene is on a plasmid or in the chromosome: use long reads or a hybrid. See plasmids and mobile genetic elements.
• Publishing a complete reference genome or genome announcement: long-read or hybrid assembly, so each chromosome and plasmid is one circular sequence.
• Comparing many strains in a pangenome: short reads for most strains, ideally with a few complete genomes as references. See pangenome analysis.

Sending data to BacSeq Analytics

Short-read data are assembled with SPAdes as part of every package. If you have long-read or hybrid data, say so when you upload your files, so the assembly approach matches the data. Whatever the strategy, the assembly is checked for contiguity, completeness and contamination before analysis, as described in assembly quality metrics.