Yes, it is common for de novo and reference-guided assemblies to yield different results. In general, if a closely related, high-quality reference genome is available, reference-guided assembly is often the best choice, especially if the goal is to obtain a complete genome or to analyze SNPs and small-scale variations. However, reference-guided assembly may miss novel sequences that are not present in the reference genome.

If no closely related reference genome is available, or if the focus is on novel sequences and evolutionary analysis, de novo assembly is preferable. But, obtaining a complete or near-complete genome requires high-quality sequencing reads.

While metrics such as N50, L50, and the number of contigs can be used for comparison, I recommend using CheckV to assess genome quality, completeness, and contamination, as well as to trim non-viral sequences. If both assemblies have similar CheckV metrics and no contamination, you can annotate them and perform sequence alignments to identify differences. Examine unique regions in the larger assembly for if they contain viral-related genes, which could indicate true biological variations rather than assembly artifacts.

Referenced assemblies can be misleading for phages, as they tend to “fill in” gaps using the guide sequence, sometimes duplicating repeats or incorporating foreign fragments. In fact, for phages such as T4, de novo assembly (e.g. with SPAdes) and not reference-based assembly is often recommended. To decide which to rely on, it is useful to compare both versions with objective metrics. For example, use QUAST/Qualimap to quantify the number of contigs, sizes (N50) and average/minimum/maximum coverage; coverage should be homogeneous across the genome, as significant variations point to contamination or partial assemblies. Repeats or duplicated regions should also be inspected: they can appear when a contig is “inflated” (dotplots with Gepard or BLASTN are useful). Specialized virus tools, such as CheckV, estimate the integrity of the viral genome and detect host contamination in a monocontig assembly.

It is also key to map the raw reads to each assembly and check the alignment: for example using BWA-MEM, Bowtie2 or Minimap2, and then inspect with SAMtools/IGV for regions with very low coverage or discordant reads. Such patterns (alteration in pair distance/orientation) usually indicate local errors. If many base errors are found, you can apply a “polish” with Pilon or another read-based polisher. To compare the two assemblies directly or with the known reference genome, you can align with MUMmer or Mauve and use dnadiff/Assemblytics to quantify structural differences. For example, BLAST or Mauve help to locate insertions, deletions or inversions in the reference-based assembly. Any additional region in the referenced assembly should be checked: does it correspond to a terminal repeat (phages like T4 have redundant ends) or is it introduced non-viral sequence? Phages can have circularly permuted genomes or terminal repeats; these can appear as “circular genome” or with internal repeats if the assembly does not handle them well. Tools such as PhageTerm (or recent end stopping protocols) serve to determine the actual end arrangement and distinguish genuine duplicates from artifacts.

In summary, beyond the final size of the assembly, rely on the one that shows the most consistency with the data and known biology. It is advisable to use multiple metrics together (not just N50): for example, assess uniform coverage in QUAST, viral integrity with CheckV, absence of foreign contigs (discard host DNA), and genomic concordance by global alignments. Recent pipeline comparisons show that, when there is a very close reference, assembling with iterative methods such as shiver or SmaltAlign can be reliable, but always validating with the above methods. Ultimately, use QUAST/Qualimap, alignment tools (MUMmer/Mauve/BLAST), read mapping and viral checks (CheckV) to detect errors; de novo assembly tuned to these checks is often more reliable. If doubts persist, consider re-sequencing (e.g., with long reads or spot PCR) to resolve discrepancies.

Greetings!

Fausto :)

Hello, Thank you both for your time and for such a thorough and helpful response. I actually combined some of your suggestions and also based my analysis on the protocol by Shen and Millard, which was extremely useful as well.

Best regards, Ana