All other positions in the peptide sequence have zero probability of being the true modification position.
Once a model is learned, we can refine the modification for each input peptide sequence by inferring its most likely PTM group, true modification mass and true modification position.
For a given peptide, the true modification position is assumed to be chosen uniformly among occurrences of that amino acid in the peptide.
The true modification position for each peptide was randomly chosen to be on one of the instances of the preassigned amino acid for that subset, and the modification positions used as input to the algorithms are set to a noisy version of the true modification positions.
The distance between the true modification position and each amino acid is used to account for our expectation that each PTM occurs on a specific set of amino acids.
We modeled the modification position error (x n − z n ) between the observed modification position x n and the true modification position z n with a discrete probability distribution, given as (4) where the likelihood function ϕ accounts for the modification position error.
Below, we described the components of our model: the probability of choosing each PTM type, the probability of choosing each amino acid to be the modified amino acid given the PTM type, the probability of the true modification position given the modified amino acid and the uncertainty in the observed modification mass and modification position.
To estimate the reliability of chromatin domain calling tools, we compared a list of true histone modification domains by different ChIP-seq peak finding algorithms.
Despite these apparent differences, our and other investigations all had few potentially susceptible individuals (< 10), making it difficult to draw conclusions regarding true effect modification.
Another interesting point is whether a continuous control of signs and symptoms in the long term translates into true disease modification or prevention of structural damage and ankylosis.
