bait design for seqCapture

[AKroxy]

thanks for your response. Yes, most people doing seqCap are using transcriptome data for bait design. But those people are interested in phylogenetics. We're interested in within-species diversity (or btw 2 very closely related species), so we don't really want conserved regions - though with 5-10,000 baits, even conserved regions would probably give enough snps to see if the 2 species are different from each other.
My husband thinks that blasting all of the unassembled reads is unreasonable bc there are millions of reads. but metagenomics people seem to do that

[nucacidhunter]

I think that targeting genic regions would be a better approach because they would be more conserved between or among species than intergenic region. To do this, assembly and a BLAST to identify contigs from coding region of genome is essential.

[AKroxy]

I have questions about developing baits for sequence capture to use in population genetics of 2 very closely related species.

We have no genomic resources for our organism (a sedge) or anything in the very large genus.

However, we did a *very* low coverage whole genome sequencing run from our species of interest on the miSeq - we ended up with about 1/10x coverage of the genome (~250bp fragments).
I want to use these unassembled reads to develop baits. We want to target random regions throughout the genome.

I had never intended to assemble this data - I didn't see a need.

After trimming, I had intended to use blast to remove as many undesirable sequences as possible (chloroplast, mitochondria, virus, bacteria, and things that blast to know repetitive DNA), and then randomly choose ~10,000 unassembled reads for bait design (designed by mycroArray/myBaits).

However, my husband, the bioinformatics guy, thinks it is crazy to design baits from unassembled sequence.

He thinks we need more sequence data, so we can at least partially assemble the genome before selecting regions for bait design. He argues that we can't trust unassembled reads because in a typical assembly (with more data) ~50% of reads can't be assembled, so they must be garbage - implying that from my approach 50% of the baits will be garbage.

I realize that we could do RADseq or some variant, but prefer not to, due to possible bias (low number of individuals per locus).

My question is whether we need to get more sequence data for a partial assembly, or if my approach is reasonable.

Thanks for any input!