ESTs are single-pass reads, so will not be a good source to find SNPs. Because you have no way to know if a variation is from sequencing error, or real SNP. Unigenes are better, as they are clusters of several ESTs, but they are still not ideal for references - because there is no guarantee that each SNP position will be covered by two or more reads.

Use genome whenever it is possible, even a draft one like papaya. The papaya genome is ~3X coverage, which means on average each position will be sequenced three times, in general lower error rate than ESTs or unigenes.

Sure. But that's not de novo assembly. De novo assembly would be using a program like velvet to build contigs without reference, then using blast to compare your contigs to your draft reference.

ESTs are going to be spliced, right? So you have short reads based on an RNA prep, not a genomic one?

Unigenes are certainly better than ESTs. I think assembled (or clustered) ESTs and unigenes can be used as reference sequences as long as you keep an eye on the variants detected during the assembling process, and you have reads from mRNAs. For reads from genomic DNAs, I think it's fine to use draft sequences as references. You'll have to verify the interesting SNPs anyway.

The question below contains some advice regarding masking the genome to contain exomes only, this relates to your question in that you have ESTs only. In general don't treat ESTs as if they were a reference genome

Exome Sequencing: Masking The Non-Genic Sequences ?

thanks all. definitely yes, my short reads are from RNA sequencing. now I'm getting more clear to choose unigene or est as a reference sequences.

I have RNAseq data of two species. Major objective is to identify SNP among both the species. strategy to find denovo SNPs could be

1. I did assembly of species 1 using ABYSS
2. mapped reads of species 2 against the contigs of species 1
3. Use samtools to find SNPs with minimum depth 10

I want to know whether it is accurate way to find denovo SNPs or not?