The classification of proteinogenic amino acids is crucial for understanding their commonalities as well as their differences to provide a hint for why life settled on the usage of precisely those amino acids. It is also crucial for predicting electrostatic, hydrophobic, stacking and other interactions, for assessing conservation in multiple alignments and many other applications. While several methods have been proposed to find “the” optimal classification, they have several shortcomings, such as the lack of efficiency and interpretability or an unnecessarily high number of discriminating features. In this study, we propose a novel method involving a repeated binary separation via a minimum amount of five features (such as hydrophobicity or volume) expressed by numerical values for amino acid characteristics. The features are extracted from the AAindex database. By simple separation at the medians, we successfully derive the five properties volume, electron–ion-interaction potential, hydrophobicity, α-helix propensity, and π-helix propensity. We extend our analysis to separations other than by the median. We further score our combinations based on how natural the separations are.

蛋白质氨基酸的分类对于理解它们的共性和差异至关重要，可以为生命为何决定使用这些氨基酸提供线索。它对于预测静电、疏水、堆积和其他相互作用，评估多重比对和许多其他应用中的守恒也至关重要。虽然已经提出了几种方法来寻找“最佳”分类，但它们有几个缺点，例如缺乏效率和可解释性或不必要的大量判别特征。在这项研究中，我们提出了一种新方法，该方法涉及通过最少数量的五个特征（例如疏水性或体积）进行重复二元分离，这些特征由氨基酸特征的数值表示。这些特征是从 AAindex 数据库中提取的。通过中位数的简单分离，我们成功地推导出了体积、电子-离子相互作用势、疏水性、α-螺旋倾向和 π-螺旋倾向五个属性。我们将分析扩展到除中位数以外的分离。我们根据分离的自然程度进一步对我们的组合进行评分。