I'm attempting to come up with a statistical model to determine the p-value for the similarity value $ 0 \le s \le 1 $ of two DNA contigs. The idea is to get a numerical measure to estimate if two contigs are homologous.

I want to account for base composition (and/or other structural features) if possible, but not necessarily because I might want to do the analysis only from the information about the length of the sequences/contigs and largest overlaps. Maybe something inspired from other statistical significance in bioinformatics .

Assume we have two DNA sequences: $a$ and $b$ with the basepair lengths $a_{\text{len}}$, $b_{\text{len}}$. Their overlap $o_{a-b}$ is the number of contiguous bases that appears in both sequences. If we define the similarity $s = \frac{o_{a-b}}{\min{(a_{\text{len}}, b_{\text{len}})}}$ how can we define a p-value for $s$?

  • $\begingroup$ Okay, can you please give some more context in your question about a specific use case for this algorithm? It will help to get better answers. $\endgroup$
    – gringer
    Jun 4 at 11:22

Well, here my thoughts:

The equation $s$ somehow tries to represent the ratio $\frac{equality}{equality+inequality}$. But $s$ returns only a value. And to assing the p-value to a single value... is complicated.

How could you approach this?

  • Using some kind of Hamming distance for each base pair at the position $n$ ($1 \leqslant n \leqslant len$) between the sequence i ($s_{i}$) and sequence j ($s_{j}$) in order to get a set of binary data:

$$s_{ij} = \begin{Bmatrix} 1\ if\ seq_{i_{n}}=seq_{j_{n}} \\ 0\ if\ seq_{i_{n}}\neq seq_{j_{n}} \end{Bmatrix}$$

With this, you get a set of binomial data (eg $\{0,1,0,1,0,1,1,1,0,1,1,1,0,...\}$). Now, you should be able to perform some (Binary?) test and get the p-value.

PD: Consider aligning the two sequences if there is interest in discussing base pair homology.

  • $\begingroup$ The overlap is what the aligner labeled as the largest similar subsequence $\endgroup$
    – 0x90
    Jun 7 at 2:06
  • $\begingroup$ Maybe something along the lines of how blast compute p-value $\endgroup$
    – 0x90
    Jun 7 at 2:24

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.