How the transition probabilities (0.02 and 0.001) were determined?

[yl-lu]

Thanks very much for the amazing software. According to the article (Cell, 2009), 'The transition probability from PrD to background was set to 0.02 and the transition probability from background to PrD state was set to 0.001. Initial probabilities for PrD and background were set to 0.05 and 0.95, respectively.'

A domain is PrD is considered as a small probability event, I think it is the reason why initial probabilities were set to 0.05 and 0.95, respectively. But I am a little confused about how the transition probabilities were decided?

[alexlancaster]

Some of this is described here: http://plaac.wi.mit.edu/details

[oliverking]

As described in the link Alex provided, those transition probabilities correspond to a Markov chain in which PrDs are expected to occur on average every 1/0.001 = 1000 AA, and have an average length of 1/0.02 = 50 AA. (Short PrDs are filtered out in a later step, and in an HMM there is also an interplay between the transition and emission probabilities that for example impacts whether two nearby PrDs will be merged into one PrD in the parsed sequence.) At equilibrium, this Markov chain is in the PrD state with probability 0.0476 (50/1050) and in the background state with probability 0.9524 (1000/1050). It is not necessary to use equilibrium frequencies as initial frequencies – one could have a model in which PrDs preferentially start at N-termini – but that's what PLAAC does. The initial frequencies in the Cell paper look to have been rounded off to the second digit.

…

On Mon, Jan 25, 2021 at 5:52 PM Alex Lancaster ***@***.***> wrote: Some of this is described here: http://plaac.wi.mit.edu/details — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#9 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABLSXDWHMDL2YVDON5VXIKDS3XYZTANCNFSM4WSOSLJQ> .