An adaptive graph learning method for automated molecular interactions and properties predictions

Improving drug discovery efficiency is a core and long-standing challenge in discovery. For this purpose, many graph learning methods have been developed to search potential candidates with fast speed low cost. In fact, the pursuit of high prediction performance on limited number datasets has crystallized their architectures hyperparameters, making them lose advantage repurposing new data generated Here we propose flexible method that can adapt any dataset make accurate predictions. The proposed employs an adaptive pipeline learn from output predictor. Without manual intervention, achieves far better all tested than traditional methods, which are based hand-designed neural other fixed items. addition, found more robust provide meaningful interpretability. Given above, serve as reliable predict molecular interactions properties adaptability, performance, robustness This work takes solid step forward purpose aiding researchers design drugs efficiency. Deep useful predictions for design, but hyperparameters need be carefully tweaked give good specific problem or dataset. Li et al. present here finds appropriate wide range tasks achieve without human intervention.