When retrieving all cds sequences from the human reference genome, in the overwhelming majority of cases, the cds is mod 3, starts with ATG, and ends in a stop codon.
However, approximately 10% are not divisible by 3 and have no stop codon (but do have a start codon in the first exon).
Are these likely to be errors?
To retrieve the sequences, I used the Table Browser, i.e.
Select -- refSeq genes for the track and ccdsInfo for the table Under output, select "selected fields from primary…" then click get output You will go to another page that gives you the option to select additional tables -- Select "ccdsGene" -- click "Allow Selection from…" That will expand another list of fields...etc.
-- click "Check all" at the top click "Get output"
You might give an example as one can come up with a number of likely explanations.
For instance, this sequence (length=119), which I obtained by selecting CDS exons with 1 FASTA record per gene:
Another example is NM_001077693 . The UCSC CDS exons give a length of 326. Moreover, when translated, there are several internal stop codons, and a frameshift difference that gives amino acid sequences inconsistent with the Genbank record for this gene.
I'm not sure what's going on with that one. If you look that up in the nucleotide database on NCBI, its sequence diverges around 1/2 way through, which suggests that something is off between NCBI and UCSC.
Look at the cdsStartStat field (hint, exon #3 is missing) in the refGene table for that entry.