Bone morphogenetic protein-2 (BMP-2) is a key regulator of bone formation, growth and regeneration, which contains a conformational wrist epitope and a linear knuckle epitope that are functionally responsible for the protein by mediating its interaction with type-I and type-II receptors, respectively. Previously, a long (19-mer) knuckle peptide derived from the knuckle epitope region (residues 73–92) has been found to promote osteogenesis and bone repair. Here, we attempt to rationally redesign the knuckle peptide by using bioinformatics and machine learning-guided evolution to obtain structurally simplified, potent osteogenic peptides that are capable of targeting type-II receptor. Complex analysis reveals that only a fraction of the epitope region can directly interact with type-II receptor, which represents a small (12-mer) knuckle-derived peptide (KDP0 peptide). Glycine scanning further identifies three KDP0 anchor residues Ser88, Leu90 and Tyr91 that are fundamentally important in the peptide–receptor binding. Systematic mutation, amino acid combination and uniform design of other nine KDP0 non-anchor residues generate 32 new knuckle-derived peptides (KDP1–KDP32); their binding affinities to recombinant protein of human type-II receptor are determined using fluorescence spectroscopy assay. The resulting affinity values (Kd) are used to train six regression models developed by combination of two machine learning methods and three amino acids descriptors. The best SVM/VHSE predictor is then utilised to guide the genetic evolution of a knuckle-derived peptide population. Eight peptides (KDP33–KDP40) with high affinity scores are selected from the improved population, and their osteogenic activities on bone marrow stromal cells are measured using alkaline phosphatase assay. Consequently, six out of the 8 tested peptides exhibit increased activity relative to KDP0 peptide. The KDP34 (DFQTWSFLYVEN) is found as the most potent peptide with APL activities of 195% and 279% at 0.01 and 0.1 µg/ml, respectively, which shares a similar binding mode with the native knuckle epitope and can form diverse nonbonded interactions of hydrogen bond, hydrophobic contact, cation-π/π-π stacking and salt bridge with type-II receptor.

骨形态发生蛋白 2 (BMP-2) 是骨形成、生长和再生的关键调节剂，它包含一个构象腕部表位和一个线性关节表位，通过介导其与 I 型和-II受体，分别。以前，已发现源自关节表位区域（残基 73-92）的长（19 聚体）关节肽可促进成骨和骨修复。在这里，我们尝试通过使用生物信息学和机器学习引导的进化来合理地重新设计关节肽，以获得能够靶向 II 型受体的结构简化、有效的成骨肽。复杂的分析表明，只有一小部分表位区域可以直接与 II 型受体相互作用，它代表一个小的（12-mer）关节衍生肽（KDP0 肽）。甘氨酸扫描进一步确定了三个 KDP0 锚定残基 Ser88、Leu90 和 Tyr91，它们在肽-受体结合中至关重要。其他九个 KDP0 非锚定残基的系统突变、氨基酸组合和统一设计产生了 32 个新的关节衍生肽（KDP1-KDP32）；使用荧光光谱法测定它们与人II型受体重组蛋白的结合亲和力。所得的亲和力值 (Kd) 用于训练通过结合两种机器学习方法和三种氨基酸描述符而开发的六个回归模型。然后利用最好的 SVM/VHSE 预测器来指导指关节衍生肽种群的遗传进化。从改良的群体中选择了八种具有高亲和力评分的肽（KDP33-KDP40），并使用碱性磷酸酶测定法测量它们对骨髓基质细胞的成骨活性。因此，8 种测试肽中有 6 种表现出相对于 KDP0 肽的活性增加。KDP34 (DFQTWSFLYVEN) 被发现是最有效的肽，在 0.01 和 0.1 µg/ml 时 APL 活性分别为 195% 和 279%，它与天然关节表位具有相似的结合模式，并且可以形成多种非键合的氢相互作用键、疏水接触、阳离子-π/π-π 堆积和与 II 型受体的盐桥。8 种测试肽中有 6 种表现出相对于 KDP0 肽的活性增加。KDP34 (DFQTWSFLYVEN) 被发现是最有效的肽，在 0.01 和 0.1 µg/ml 时 APL 活性分别为 195% 和 279%，它与天然关节表位具有相似的结合模式，并且可以形成多种非键合的氢相互作用键、疏水接触、阳离子-π/π-π 堆积和与 II 型受体的盐桥。8 种测试肽中有 6 种表现出相对于 KDP0 肽的活性增加。KDP34 (DFQTWSFLYVEN) 被发现是最有效的肽，在 0.01 和 0.1 µg/ml 时 APL 活性分别为 195% 和 279%，它与天然关节表位具有相似的结合模式，并且可以形成多种非键合的氢相互作用键、疏水接触、阳离子-π/π-π 堆积和与 II 型受体的盐桥。