@techreport{oai:ipsj.ixsq.nii.ac.jp:00197714,
 author = {李, 佳男 and 吉川, 寧 and 大上, 雅史 and 秋山, 泰 and Jianan, Li and Yasushi, Yoshikawa and Masahito, Ohue and Yutaka, Akiyama},
 issue = {52},
 month = {Jun},
 note = {環状ペプチド医薬品は従来のペプチド医薬品と異なる大環状構造を持ち，低分子医薬品と比べて標的に対する特異性が高く，さらに経口投与が可能なものもあるため注目されている．体内安定性は医薬品開発の重要な指標の1つであり，全身循環血中に到達した医薬品がどの程度安定に存在できるかを表し，血漿タンパク質結合率（PPB）と密接な関係がある．環状ペプチドの血漿タンパク質結合率に関する先行研究により，局所構造が血漿タンパク質結合率に大きな影響を与えることがわかっている．しかし，現在よく使われている特徴量計算ソフトウェアは低分子化合物用に設計され，化合物全体の構造から特徴量を求めており，局所構造の検討が難しい．我々はこれまでに，汎化性能の高い特徴量を用いて低分子化合物で予測モデルを構築し，環状ペプチドの血漿タンパク質結合率を予測した．しかし，その予測精度は，実用に十分といえるレベルではなかった．そのため，本研究は環状ペプチドを残基単位で分割し，残基から計算された特徴量を加え，環状ペプチドの血漿タンパク質結合率予測手法を改良した．その結果，訓練データの交差検証では実験値と高い相関のある予測結果（R=0.90）を得ることに成功した．さらに，独立した検証データに対し，実験値と良い相関のある予測結果（R=0.83）を得た．, Cyclic peptide drugs are attracting attention because they have macrocyclic structures that are different from conventional peptides, they have high target specificity compared with small molecule drugs, and some can be orally administered. The internal stability of drugs is an important indicator for drug development, indicates how stably the drug that has reached the systemic circulation can exist, and is closely related to the plasma protein binding (PPB). Previous studies of PPB of cyclic peptides have shown that local structure has a significant effect on PPB. However, currently used descriptor calculation software is designed for small molecule compounds. The descriptor is obtained from the structure of the whole compound, and it is difficult to reflect local structure. We previously constructed a PPB prediction model for cyclic peptides with small molecule compounds using descriptors with high generalization performance. But the prediction accuracy was low and it was difficult to put to practical use. Therefore, in this study, the cyclic peptides were separated residue by residue, and the descriptors calculated from each residue was added to improve the method for predicting PPB of cyclic peptides. As a result, in cross-validation of training data, we succeeded in obtaining a prediction result with high correlation with experimental values (R = 0.90). Furthermore, for an independent test set, we obtained prediction results with good correlation with experimental values (R = 0.83).},
 title = {機械学習を用いた環状ペプチドの体内安定性予測手法の改良},
 year = {2019}
}