Discovery of hemostatic component combination from Nelumbinis Receptaculum using dual machine learning...

Nelumbinis Receptaculum (NR) were hemostatic herbal medicines for treating metrorrhagia, hematuria or hemorrhoids in ethnopharmacological systems, such as Ayurveda and traditional Chinese medicines.

The material basis for traditional hemostatic efficacy of NR is unclear. Thus, hemostatic component combination (HCC) of NR were screened by dual machine learning-based spectrum-effect relationships (DML-SER) analysis.

Ultra-high performance liquid chromatography tandem with mass spectrometry and simvastatin-induced intestinal bleeding zebrafish models were respectively used for chemical profiling and hemostatic evaluation of NR extract. Grey relational analysis were used to preliminarily correlate quantified compounds with hemostatic efficacy. Dual machine learning of artificial neural network (ANN)/support vector regression (SVR) were used for modeling spectrum-effect relationships between multi-compound contents in NR extract and zebrafish intestinal bleeding incidences. Potential HCCs were designed by compounds with top-10 permutation importance (PI) and verified by comparing hemostatic indicators with NR extract on zebrafish.

Totally, 266 compounds were characterized in NR extract, among which 42 were quantified. All NR extract showed hemostasis with decreased intestinal bleeding incidences on zebrafish. Grey correlation degrees of quantified compounds were 0.670-0.854, suggesting their relevance to hemostasis. Dual machine learning of ANN/SVR were more appropriate than partial least squares regression for modeling spectrum-effect relationships of NR extract with smaller mean absolute errors and root mean square errors but larger coefficients of determination (R² = 0.939/0.981). Combination of intersection for ANN/SVR top-10-PI compounds, containing gallocatechin, catechin, nuciferine, hyperoside, isoquercitrin and quercetin-3-O-glucuronide, exerted similar hemostatic indicators on zebrafish with whole NR extract.

Using DML-SER, a potential HCC containing six compounds were discovered and initially verified, which could provide a material basis to facilitate mechanism studies, quality control and clinical application of NR.