After each iteration, the matrix M is a better estimation of the common segments in all the sequences because their distribution is similar (the matrix will be motif biased) while the rest of segments are different and their distributions are therefore contributing in a little amount to the construction of the matrix.