全網羅ミルクペプチドアレイを用いたアレルギー診断ペプチド探索への分布関数解析法の応用 Comprehensive Combination Analysis for Screening of Significant Peptide Epitopes Using a Slide Glass Type-Exclusive Peptide Array from Milk Protein

ミルクアレルギー発症は血清中のIgEが認識するエピトープの位置によるという仮説のもと，診断に重要なIgEエピトープを，分布関数を用いた網羅的組み合わせ解析法（CC-GAM）で解析した．弱識別器として２分枝の決定木を用い，Adaboostアルゴリズムにより3入力全網羅で600万種類以上のモデリングを実施し，10-fold CVでの正答率の頻度分布を描出した．EMアルゴリズムで解析したところ，総合正規分布は，ベイズ情報量基準から分布数5個の混合正規分布で表現できることがわかった．基準正答率より高い平均値をもつ“重要正規分布”に帰属する識別器の入力変数を解析したところ，帰属頻度の高い重要なペプチドエピトープ16種類が特定できた．aS1C30，aS1C47は同一タンパク質内組み合わせ解析でも選別された．特にaS1C30の一部（SERYLGYL）は既往の研究でも重要なペプチドとして報告されていることから，我々の選択方法が診断に重要なペプチドを選択する方法として適していることを示唆した．

Based on the hypothesis that development of cow's milk allergy (CMA) in children can diagnosed from species of IgE epitope in patients, significant peptide epitope for IgE was analyzed by comprehensive combination analysis using the graphic assessment method. The Adaboost algorithm was used for classification of decision stump as weak classifier, and 6044060 models were constructed. Classification accuracies by 10-fol CV (10-fold cross-validation) were determined. Total distribution of accuracies was depicted and it was analyzed by Gaussian mixture distribution using the EM algorithm. It was found that the distribution could be divided into 5 sub-distributions by the Bayesian information criterion. The Gaussian sub-distributions were grouped into “significant sub-distributions” and “non-significant sub-distributions” and classifiers that belong to “significant distribution” were identified. Analysis of the assignment frequency of individual inputs assigned to the “significant distribution” identified 16 peptide epitopes with high assignment frequency, including aS1C30 and aS1C47 peptides. Since SERYLGYL from aS1C30 has already been reported as significant peptide epitope for CMA diagnose, our proposed method, CC-GAM, was suggested to be a useful tool for selection of significant epitopes.