This service can generate descriptors from protein files and ligand files for modelling.
Introduction to this website
Virtual screening (VS) based on molecular docking has emerged as one of the mainstream technologies of drug design and has contributed a lot to the discovery and development of a variety of new drugs due to its low cost and high efficiency. However, scoring functions (SFs) implemented in most docking programs are not always accurate enough. Therefore we propose an integrated platform called ASFP, which aims to develop customized high-precision generic or target-specific SFs for binding affinity prediction and structure-based VS through artificial intelligence (AI) technologies.
In this section, 15 computational tools were embedded into the module for the characterization of protein-ligand interaction information, and they can provide up to 3437 descriptors.
Among them, DPOCKET can compute descriptors for protein binding pockets; ECFP and PaDEL can generate structural descriptors for ligands and the others are the energy terms computed by existing scoring functions to represent protein-ligand interactions.
In view of copyright issues with some software, we have stopped external services.
In this section, 6 computational tools provide a total of 384 descriptors. The dataset is divided into the training set and the test set and will be preprocessed using sklearn.
Three machine learning algorithms (Random Forest, Support Vector Machine and eXtreme Gradient Boosting) are provided. Users can decide to use which ML algorithm and do settings about hyperparameter optimization (which hyperparameter to be optimized, the hyperparameter range and the optimization times). Finally, according to the user's input, the server uses hyperopt to find the optimal hyperparameter combinations and chooses the corresponding ML algorithm for training (10-fold cross validation) and prediction, and output the results into a PDF file.
In this section, we provide 15 well-prepared target-specific classification models for the virtual screening of those corresponding targets, and a generic AI regression model for binding affinity online prediction. Based on the input files, the descriptors were computed by the 6 preset computational tools (AffiScore version 3.0, AutoDock version 6.8, DSX version 0.9, GalaxyDockBP2, NNScore version 2.01 and SMoG2016) and then the results were predicted by the established model chosen by the user. The output of the results is shown in a PDF file.
This website mainly provides the following functions