Prediction of protein domain boundaries from statistics of appearance of amino acid residues