Diving, and even digging, into the wild jungle of annotation pathways for non-vertebrate animals

This new version of the manuscript is enriched with corrections and precisions to anwered me and other reviewers questions and suggestions, making it a better article.

The authors have provided an answer to all my main comments, and I mostly agree with them. I still believe that a rough estimate of the current (and past) costs for different techniques could be provided, acknowledging that this may become quickly obsolete. This would provide an upper range estimate for teams intending to start a genome sequencing project. The review is already thorough and I am happy to support acceptance.
Best wishes,

Yann Bourgeois

The authors have successfully addressed all my concerns.

First I would like to thanks the authors for the work they have done. Here they present a review paper about sequencing non-vertebrates genomes. As a whole, this paper is very pleasant to read.

Each part is rich of details on history of technologies and methods. Presentation of tools is quite exhaustive. Those two arguments made this paper an excellent starting point for non familiar people with sequencing technologies and more particularly for sequencing non-vertebrates genomes.

In figure 2, I would recommand to use some color to make the message easier to understand, and to use a monospace police for the consensus part.
The central part of the figure 5 might be improved, maybe with clear arrows direction and starting point.

I read the manuscript titled «A deep dive into genome assemblies of non-vertebrate animals» by Guiglielmoni et al. with great interest. The authors talk about existing methods and algorithms for constructing contiguous and accurate genome assemblies in the context of metazoan genomes. In my opinion, the article is well written and easily understandable by non-specialists. I only have minor concerns that I would like the authors to address if they agree with me.

## Introduction
7;894 => 7,894

## Sequencing
Figure 1: I understand the intent of this figure, but I find it pretty challenging to read, and points hide other points. One way of fixing this would be to aggregate the data of each category per year and turn it into a boxplot.
«The resulting reads have a length around twenty kilobases (kb)»: In my experience, PacBio reads usually have a mean size around 15kb that can go up to 25kb (see https://www.nature.com/articles/s41597-020-00743-4 as an example).
«The error rate has also been decreasing with the release of new flow cells and the development of more accurate basecallers such as Bonito.» There is also a new protocol called Q20+, which makes it possible to generate reads with a 1% error rate.

## Genome assembly
«DBG-based assemblers require highly accurate reads in which errors are only substitutions, with no indels»: why should there be no indels?
«To this end, heterozygous regions are collapsed in order to keep a single sequence for every region in the genome»: this is true if the genome is not very heterozygous. In the other scenario, both haplotypes can often be retrieved, as heterozygous regions are pretty different.

## Assembly pre and post-processing
Table 2 - Long reads error correction: NaS is missing. https://bmcgenomics.biomedcentral.com/articles/10.1186/s12864-015-1519-z
Table 2 - Short and long reads polishing: a new tool called HAPO-G has been published recently and is absent from the list. It has been developed explicitly to polish heterozygous genomes but also handles homozygous genomes. https://academic.oup.com/nargab/article/3/2/lqab034/6262629
Figure 6: Same as Figure 1
Drawbacks of using Hi-C are not presented. As an example, the fact that gap sizes cannot be estimated is not indicated.
«Assembly and pre/post-processing steps are often combined in one tool» makes it look like there is no need to post-process assemblies further, but if the polishing step is only done with long reads, the final quality will not be great.

## Phasing assemblies
Hifiasm is another assembler that can phase haplotypes.

This work reviews the current state of methods for genome sequencing and de novo assembly, with a particular focus on invertebrates, for which resources are still missing. This sort of work should be encouraged, as it aims at expanding genomic resources to non-model species, which is crucial to obtain a more comprehensive picture of the evolutionary and mechanistic processes underlying biological diversity. The “technical” content is comprehensive and mostly up-to-date. My main concerns are mostly revolving around the structure and the scope of the review. In its current state, it reads like a rather “generic” review about assembly tools, with illustrations drawn from genomic studies of invertebrate species. I think that the review would benefit from a more explicit description of the specific challenges encountered in invertebrates. Low DNA amounts is mentioned, but there are other aspects that could be described. For example, many species are difficult to raise in controlled conditions, or rare in the wild, or poorly described from a taxonomic perspective. On the other hand, many species of arthropods reproduce asexually (e.g. Daphnia), which may help increasing the yield of DNA from the same genotype. At the moment, it reads more like a collection of anecdotes (which I agree all reveal an interesting problem): there may be a better way to structure it.

It would also be good to explain from the beginning the readership that this review targets. For example, I understand the interest of adopting a historical perspective in the first section (Sequencing) if the review is a resource for new practitioners. However, a review that aims at explaining the current methods for genome assembly to "naive" readers should take more time explaining basic concepts (e.g. N50). A glossary could be useful. On the other hand, if the review is addressed to scientists who already have some experience with the techniques and the terms, the somewhat long description of Sanger sequencing may not be particularly useful.

In my opinion the review does not provide (yet) a guide to decide of a sequencing strategy. The information is already there, but could be highlighted in a more organized way. Figure 6 is a good example of what could be done more extensively throughout the review in my opinion (with more details).

The authors could compare the quality of currently available assemblies, using several metrics, and highlight the methods used to obtain them. For example, what sequencing depth of coverage is needed when using only Illumina reads + mate pairs? Hi-C? PacBio + Illumina short reads? What is the average cost? It would be useful to have figures such as decision-making flowcharts. Figure 5 could be expanded to highlight the different possible options at each step (short-reads? Long-reads? What is the best option given a budget of 10,000$? 50,000$?). What are the bioinformatic resources needed? What is the runtime of different programs, and how this runtime scales with genome size and complexity?

I also think that mentioning reference-guided assemblies could be useful, especially for readers who consider working on a species related to one that has already been sequenced. If there are reasons to assume that synteny is high and divergence low, reference-guided assemblies may be a good way for researchers with limited financial resources to obtain a valuable resource. A particularly interesting paper from this perspective (in my opinion) is the following one (Lischer & Shimizu, BMC Bioinformatics, 2017):

https://bmcbioinformatics.biomedcentral.com/articles/10.1186/s12859-017-1911-6

Note that this paper also proposes an interesting way to test for the quality of assemblies obtained by different methods through the combination of 36 summary statistics (using z-scores for each of the statistics and comparing their distributions across methods).

At last, it may be worth explaining what can be done with a genome assembly depending on its quality. If the goal consists in running preliminary population genetics analyses, a fragmented assembly can already be very useful. For comparative genomic analyses, assessment of repetitive content (transposable elements), or functional studies, high quality assemblies are the target to reach.

Nevertheless, I want to emphasize the fact that the review is rather comprehensive, and mostly needs polishing to increase its impact on a broad range of readers.

Minor comments through the text:

Introduction, Paragraph 6: The bit about BUSCO feels slightly too long, although the issue highlighted is very interesting. There are many other possible biases that could be discussed. Maybe shorten it, and provide other examples of how bias towards model systems can impair research on non-vertebrates. In general, the Introduction would benefit from explicitly stating the scope of the review, and what it means to achieve (decision-making tool? Comparison of methods? Introduction to the field for new practitioners?).

Sequencing, second paragraph. N50 is usually low for second generation sequencing, as you mention, but using Hi-C, Hi-Fi or mate pairs (which I would still classify as second-generation sequencing) can improve assemblies a lot.

Sequencing, third paragraph. The current increase in accuracy for base calling and assembly from nanopore reads is encouraging, but should be discussed more in terms of minimum depth of coverage required, the quality of training datasets (for algorithms using machine/deep learning), etc. Note the existence of another base-caller, Poreover, to be used in combination with Bonito https://github.com/jordisr/poreover

Table 1: This table is a good resource, but it may be worth considering merging it with table 2. A classification highlighting speed and memory requirements would be useful. As mentioned in the main comments, I am not sure that the row on first-generation sequencing is particularly useful.

P8: You talk here about k-mers, but what about decisions on which k-mer length to use? Why is it important to use several k-mer lengths when assembling? This is something that you could already explain here.

P17: Assembly evaluation. There are so many ways to estimate the quality of an assembly that some authors have proposed a set of tens of summary statistics, that they summarize as a Z-score. (check papers on reference-guided assemblies).

Figure 4: It would be interesting from a decision-making perspective to add a panel with the different techniques used to assemble these genomes.