Does anyone know if there is an established convention or rule on the minimum percent identity cutoff for nucleotide sequence in order to infer sequence homology?
I am looking for pseudogenes and have found regions that show ~45% identity to the exons of my gene. Not to mention that they are in the correct order and syntenic region…
But presumably, if we put in enough spaces, we can align any two sequences... Is there minimum identity value where one can be confident that they are looking at truly homologous regions?
Any advice/opinion would be greatly appreciated!
A 30% identity threshold for homology underestimates the number of homologs detected by sequence similarity between humans and yeast by 33% (this is a minimum estimate; even more homologs can be detected by more sensitive comparison methods). The bit-score provides a better rule-of-thumb for inferring homology.
Hi..
I think 200bp at least for sequence, if you need for homology, you must use phylogenetic tree analysis programs that can identified any homology.
Hello,
I liked the paper attached by Patrick SENAM KOFI Fatsi. You may find your answer in that paper. I have a couple of comments:
This is not surprising that pseudogenes show identity to the exons of your gene of interest. The pseudogene, at some point may be a functional gene, that was no longer required after some time and therefore accumulated mutations and lost its functioning. They may be in the process of "pseudogenization", I use this word, I am not sure if it exists!
Second, yes, theoretically you can align any two sequences, if you are only aligning them, without worrying about the sequence similarity or identity. But in practice, when the alignment is right, you can see conserved and diverged regions clearly, even in 45% identity.
Best,
S
If you go back to the paper suggested by Patrick SENAM KOFI Fatsi, you will see that they suggest using e-value rather that %identity. The e-value is a much better measure of statistical significance of your homology. For example, you can easily find a 90% nucleotide identity for 15 bases just by chance (and thus get a relatively high e-value, i.e. an e-value indicating no statistically significant relationship), but it is very unlikely that you will find 70% nucleotide identity in a sequence >1000 bases (which would probably give you an e-value smaller than 10^-70).
However, with regards to the 30% identity mentioned before, reading the paper it is my understanding that it refers to protein homology, where much lower identity is considered statistically significant (due to 20 amino acids, vs 4 bases).
What is the correct percentage of sequence identity for finding true homologous genes?
It is possible to address this question with a simple experiment. One can take two homologous nucleotide sequences with low % nucleotide identity such as 50% and align them using different algorithms. One algorithm, for example, Geneious alignment (Biomatters), can deliver an alignment with 50% identity. ClustalW tool will give an alignment with 45% identity for the same two sequences. The latter will involve entirely different regions compared with the 50% identity alignment. This may indicate that % nucleotide identity as low as 45 is associated with false alignments that do not reflect the homology. This can be studied at a high scale with dedicated bioinformatic tools and adequate statistical support to find a precise threshold, which will possibly be lineage-specific. At the same time, there can be sequences with multiple regions of homology due to duplication events. Such sequences can seriously compromise the experimental determination of the true biologically relevant threshold of % nucleotide identity for homologous sequences.
If one builds pairwise alignments for a large subset of randomly selected cDNA sequences, it will be obvious that % nucleotide identity values between 49 and 55 come just by chance. Thus, there is no need to refer to 50% as "low", contrary to my reply above. 50% is rather a baseline that has nothing to do with true homology. In fact, it is rather difficult to find alignments with values below 49% in a random sample of alignments. Sometimes, if only an end of one sequence is involved in the alignment, much higher % identity values (up to 60 or more) can be found in the sample of random alignments. Thus, % identity alone cannot serve as a reliable indicator of homology when it comes to nucleotide alignments unless the values are really large. However, values as low as 57% nucleotide identity can be associated with true homology. These conclusions were made based on the Geneious alignment tool run on plant transcripts (Medicago truncatula).
- Badges
- Science topic
- Similar topics
- Geoscience
- Asia
- India