Difference between assembly, gene annotation, and reference genome?

[prs321]

What are the differences? Why make an assembly when you can just map reads to the reference genome and look for mutations that way?

[GenoMax]

That is only possible when a "reference" genome is available. Many do not work with model/well-annotated genomes/organisms.

If one generates a "great" assembly then it can eventually lead to a "reference genome" which can ultimately be "annotated".

At that point someone else can come along and do what you proposed above (i.e. map and look for mutations that can be functionally mapped).

[prs321]

So what is considered a "great" assembly?

Is an assembly constructed by mapping a sample over an available reference and then assembling the sequences into a "better" reference genome? Thus ending up with a better reference genome aka the assembly?

[GenoMax]

So what is considered a "great" assembly?

There are bound to be differences of opinion about the definition but something that reasonably accounts for/characterizes a "genome" can be considered one.

Is an assembly constructed by mapping a sample over an available reference and then assembling the sequences into a "better" reference genome? Thus ending up with a better reference genome aka the assembly?

Here is a quote from the GRC's main web page that addresses this question.

The original model for representing the genome assemblies was to use a single, preferred tiling path to produce a single consensus representation of the genome. Subsequent analysis has shown that for most mammalian genomes a single tiling path is insufficient to represent a genome in regions with complex allelic diversity. The GRC is now working to create assemblies that better represent this diversity and provide more robust substrates for genome analysis.

[Wallysb01]

So what is considered a "great" assembly?

As Genomax said, there will be a variety of definition of "great", but I think you can break down useful genome assemblies into about 3 catagories of increasing "greatness":

1) scaffold N50 roughly equals the genome size of a single gene for that organism (for vertebrates this would be roughly 50kbp. Due to the fragmented nature of the genome, hopefully you have a very low percent of Ns in there as well. This would be the kind of genome created from high coverage illumina data or low coverage sanger, but without long scaffolding libraries or physical mapping of contigs/scaffolds to chromosomes.

For this kind of genome, you could at least find many mostly whole genes to some preliminary work, but it will be difficult to do many of the genomic studies possible in mice/human. Here the sheer number of scaffolds/contigs in your assembly will prove challanging to analyze, as this number can be in the hundreds of thousands.

2) scaffold N50 increases to include many genes. This would be in the >1Mbp or greater range for vertebrates. For this type of genome, you probably have a lot of gaps, but hopefully your contig N50 is >10Kbp. Now, you'll have synteny blocks. You can start doing more advanced forms of genomic analysis due to the fewer number of contigs/scaffolds (maybe in the low thousands for vertebrates).

3) scaffolds are now physically mapped to chromosomes, allowing for scaffold N50 to approach the size of small chromosomes (say 50Mbps) and total number of scaffolds approaches the haploid chromosome count for the species. Also, gaps in the genome are systamtically filled in so N% is approaching zero. This has really only been mostly completed, or even attempted, for a handful of genomes to date.

Is an assembly constructed by mapping a sample over an available reference and then assembling the sequences into a "better" reference genome? Thus ending up with a better reference genome aka the assembly?

Most genomes are completely de novo, which allows for not biasing your genome towards what ever structure is in the nearest reference genome. But if you're working on a very closely related species with a reference, or maybe a bunch of them, and you don't care about getting the genomes of most them all that correct, this can be a cost effective way to make a genome. Particularly, if you're interested in just presence/absence of SNPs or small indels. But those kinds of genomes come with a lot of caviates, since without a de novo genome assembly its pretty difficult to know just how wrong your reference guided genome assembly might be.