[关键词]
[摘要]
目的 针对传统网络药理学固定阈值及输入成分构成易导致结果不稳定的问题,构建参数与成分双扰动鲁棒性框架,将单一条件下的靶点筛选转化为跨参数、跨成分的稳定性排序,并以筛选草莓叶抗氧化活性成分为例示范应用。方法 基于文献报道的液相色谱-串联质谱(liquid chromatography-tandem mass spectrometry,LC-MS/MS)注释结果,经结构规范化和去重获得18个化合物;采用SwissTargetPrediction预测靶点,整合GeneCards、OMIM和TTD构建氧化应激相关基因集合。围绕靶点预测概率、STRING互作置信度和按节点度值排序选取前K个枢纽靶点(Hub Top K)开展112组全因子参数扰动,并逐一剔除18个化合物,完成2 016组成分扰动运行。依据跨参数出现频率、靶点平均归一化度值(degree)或功能条目平均富集强度筛选参数稳健结果;再根据成分剔除保留率确定基于预设操作性阈值的双重稳健候选结果;结合反向药效团匹配靶点预测平台(PharmMapper)和相似性集成方法(similarity ensemble approach,SEA)跨平台靶点预测、PubMed共现、节点度匹配置换检验和重复分子对接进行评估。结果 共获得684个非重复预测靶点,其中286个预测概率大于0的靶点与氧化应激相关基因存在交集。候选靶点和功能富集结果随筛选阈值变化而波动。最终获得50个参数稳健靶点和70个参数稳健功能条目,其中29个靶点和58个功能条目依据预设操作性阈值,在成分扰动后被判定为具有相对稳健性。PharmMapper和SEA分别为13个和5个基于预设操作性阈值判定的相对稳健候选靶点提供补充性计算支持,其中碳酸酐酶2(carbonic anhydrase 2,CA2)和CA12同时获得3个靶点预测平台的支持。信号转导和转录激活因子3(signal transducer and activator of transcription 3,STAT3)在全成分条件下的参数出现频率为100%,且在全部成分剔除条件下均保持参数稳健状态;表皮生长因子受体(epidermal growth factor receptor,EGFR)、酪氨酸激酶Src(proto-oncogene tyrosine-protein kinase Src,SRC)、肿瘤坏死因子(tumor necrosis factor,TNF)、热休克蛋白90α家族A类成员1(heat shock protein 90 alpha family class A member 1,HSP90AA1)和B淋巴细胞瘤-2(B-cell lymphoma-2,BCL2)等亦具有较高稳定性。稳健功能条目主要涉及氧化应激响应、炎症调控、细胞凋亡和激酶信号转导等共享功能模块。结论 参数与成分双扰动框架可同时评估人为阈值和输入成分依赖性,从而提高网络药理学候选结果的稳定性与可审计性。
[Key word]
[Abstract]
Objective To address the instability of conventional network pharmacology analyses arising from fixed screening thresholds and dependence on input compound composition, a dual-perturbation robustness framework was constructed to transform the target screening under a single condition into stability ranking across parameters and components. The application was demonstrated by taking the screening of antioxidant active components from Caomeiye (Folia Fragariae) as an example. Methods Eighteen compounds were retained after structural standardization and deduplication of previously reported liquid chromatography-tandem mass spectrometry (LC-MS/MS) annotations. Potential targets were predicted using SwissTargetPrediction, and an oxidative stress-related gene set was compiled from GeneCards, OMIM, and TTD. Based on the predicted probability of the target, the confidence of STRING interaction, and the ranking by node degree value, the Top K Hub targets (Hub Top K) were selected to carry out 112 sets of full-factor parameter perturbations. Each of the 18 compounds was then excluded in turn, resulting in 2 016 compound-perturbation runs. The robust results of the parameters were screened based on the cross-parameter occurrence frequency, the average normalization degree value (degree) of the target, or the average enrichment intensity of the functional items. Dual-robust candidates were subsequently defined based on retention rates following compound exclusion under prespecified operational thresholds. Complementary evaluations included cross-platform target prediction using PharmMapper and similarity ensemble approach (SEA), PubMed co-occurrence analysis, degree-matched permutation testing, and repeated molecular docking. Results A total of 684 nonredundant predicted targets were obtained, 286 with positive prediction probabilities overlapped the oxidative stress-related gene set. Candidate targets and functional enrichment results varied across screening thresholds. A total of 50 parameter-robust targets and 70 parameter-robust functional terms were identified, of which 29 targets and 58 functional terms were classified as relatively robust based on prespecified operational thresholds after compound perturbation. PharmMapper and SEA provided complementary computational support for 13 and five relatively robust candidate targets identified based on prespecified operational thresholds, respectively, carbonic anhydrase 2 (CA2) and CA12 were identified by all three target-prediction platforms. Under the full-compound condition, the parameter occurrence frequency of signal transducer and activator of transcription 3 (STAT3) was 100%, and the parameter maintained robust status under every compound-exclusion condition. Epidermal growth factor receptor (EGFR), proto-oncogene tyrosine-protein kinase Src (SRC), tumor necrosis factor (TNF), heat shock protein 90 alpha family class A member 1 (HSP90AA1), and B-cell lymphoma-2 (BCL2) also showed high stability. The robust functional terms primarily represented shared modules related to oxidative stress responses, inflammatory regulation, apoptosis, and kinase signaling. Conclusion The parameter and component dual-perturbation framework can simultaneously evaluate artificial thresholds and input component dependencies, thereby improving the stability and auditability of candidate prioritization in network pharmacology.
[中图分类号]
R285
[基金项目]
北京市农林科学院科技创新能力建设专项(KJCX20261801,KJCX20261003)