The integration of multiple virtual screening strategies facilitates the balance of computational efficiency and prediction accuracy. In this study, we constructed an efficient and reliable “multi-stage virtual screening-in vitro biological validation” system to identify potential inhibitors targeting extracellular signal-regulated protein kinase 2 (ERK2). Firstly, we rapidly obtained 10 candidate ERK2 inhibitors with desirable pharmacokinetic characteristics from thousands of named natural products in ZINC database based on machine learning classification models and ADME/T prediction. The structure-based molecular docking approach was then used to obtain four further hits with lower binding free energy compared to the positive control molecule Magnolipin. Subsequently, the two compounds were purchased for in vitro biological validation considering commercial availability and economic cost, and the results showed that Dodoviscin A exhibited acceptable inhibitory activity on ERK2 (IC50 = 10.79 μm). Finally, the mechanism of action and binding stability of this natural product inhibitor were investigated by binding mode analysis and molecular dynamics simulation.

Nowadays more and more people like to invest in volatile assets, and it is the goal of every market trader to maximize the total return by developing a reasonable investment strategy. We first predicted the daily value of gold and bitcoin for five years based on known data, we built two models, one is Improved Metabolic Gray Model (Abbreviated as IGM), the other is Time Series Model ARIMA. The application of the model helps investors make investment decisions and improve economic returns.

Nuclear receptors (NRs) play a crucial role in the pathogenesis of metabolic syndrome. Farnesol X receptor (FXR) and retinoid X receptor (RXR) are members of the NR superfamily and are usually present as heterodimers in vivo. Screening for multi‐target NR activators is of great importance due to the complex pathogenesis of metabolic diseases. Virtual screening is often used for drug discovery. In this study, we first collected data on relevant compounds and subsequently constructed three machine learning models (Random Forest, Support Vector Machine, and Artificial Neural Network) with molecular descriptors. We then performed model fusion based on the soft voting strategy and the prediction accuracies of the fused models were 85.2 % for the FXR external validation set and 84.3 % for the RXRα external validation set, followed by virtual screening of compounds in the ZINC database according to a model score threshold of 0.7. The 499 hits commonly selected by both models were then subjected to a drug‐likeness filter and ADMET prediction using appropriate bioavailability parameters. The 10 compounds that met the criteria were processed by molecular docking using the AutoDock Vina software. The results showed that we screened four potential dual FXR/RXRα agonists with binding energies less than −8.0 kcal/mol for both targets (ZINC1557163, ZINC14824986, ZINC5273978 and ZINC40394465).