Contigs Vs Scaffolds for Assembly Analysis

[narain]

Hi Members

I am using Assembly tools to generate contigs and scaffolds. One tool gives contigs with higher N50 than another tool. The same tool however gives scaffolds with lower N50 than the other tool. Should I use contigs or the scaffolds to conduct analysis such as comparing with the reference genome ? Scaffolds are essentially set of contigs with estimates of gaps in between which are represented by Ns.

Aby

[tonybolger]

I am using Assembly tools to generate contigs and scaffolds. One tool gives contigs with higher N50 than another tool. The same tool however gives scaffolds with lower N50 than the other tool. Should I use contigs or the scaffolds to conduct analysis such as comparing with the reference genome ?

Scaffolds, if you can. Not all tools work well with long N stretches though.

[narain]

Hi Tony

Thank you for your suggestion for using scaffolds. But the question reduces further to scafoold from which tool will be more sensible.

As I said earlier, the scaffolds generated by 1st tool have less N50 than the scaffolds generated by 2nd tool. However, the contigs in the scaffolds of the 1st tool have higher N50 thank the contigs in the scaffolds by 2nd tool. Does this not mean that the higher N50 in the scaffolds generated by 2nd tool is all attributed to more N's in between, and thus the scaffolds of the 2nd tool are not as reliable as the scaffolds of the 1st tool, though the second tools scaffolds have higher N50 ?

Aby

[tonybolger]

Thank you for your suggestion for using scaffolds. But the question reduces further to scafoold from which tool will be more sensible.

As I said earlier, the scaffolds generated by 1st tool have less N50 than the scaffolds generated by 2nd tool. However, the contigs in the scaffolds of the 1st tool have higher N50 thank the contigs in the scaffolds by 2nd tool. Does this not mean that the higher N50 in the scaffolds generated by 2nd tool is all attributed to more N's in between, and thus the scaffolds of the 2nd tool are not as reliable as the scaffolds of the 1st tool, though the second tools scaffolds have higher N50 ?

I would compare the two sets of scaffolds. You need to make a combined judgement based on N50/total size and other factors, such as how well the assembled sequences seem to match some reference or part thereof.

N50 isn't really a definitive measure of quality - it can just indicate that one tool is taking more risks than the other, which considers a particular joining as insufficiently supported. You can get a wonderful N50 by just concatenating all the reads - it's not likely to be correct though.

I wouldn't really worry about the contig vs scaffold size strangeness - some tools apply some approaches or data later than others. For example, SOAPdenovo doesn't seem to use pairing information at all at the contig stage, so its contigs are always relatively small vs alternatives which do use pairing, but on the other hand, it maintains a full graph for scaffolding, so it works much better with its own contigs than an imported set.

[narain]

Hi Tony

Thank you for suggestion. It looks quite tedious task to compare the quality of the assembly then if not just a parameter such as N50. Thanks for those inputs though I feel the N50 which has been fairly established since many years should also be kept as one of the prominent criteria to judge. I wasn't naming the tools intentionally as then members will get biased in their replies and comments.

Aby

[tonybolger]

Thank you for suggestion. It looks quite tedious task to compare the quality of the assembly then if not just a parameter such as N50.

Agreed. There are some quality checking approaches out there (e.g. from AMOS) which give some numbers - not sure if they'll scale to your dataset.

Thanks for those inputs though I feel the N50 which has been fairly established since many years should also be kept as one of the prominent criteria to judge.

I agree - though N50 must be combined with total size or % completeness. Otherwise you can just game the N50 by dropping all of the short contigs/scaffolds.