Src kinase app: valid inhibitor generation and prediction with explanation using predictive model and selfies

Using the predictive model to a virtual screen of a large data set and feeding it to a Recurrent Neural Network using SELFIES is a new way to generate valid and active molecules without the need or guide of optimization through reinforcement learning, and also provides a place to save those molecules for free and provide a virtual screening app to predict Src kinase activity and using Explainable Ai to understand what model do. In this study, the author train the modern Artificial Intelligence model including Machine learning, Deep learning and validates it using 50k random curated Zinc compounds and gives the result of docking range and residue of each model to make an example of the ability of each model, and also make a virtual screen by Artificial Intelligence model to 3 million from ZINC database with 500k ChEMBL compounds and feed the most active to Recurrent Neural network using SELFIES and generate 100 compounds for each Temperature and perform structure-based docking and protein-ligand interaction, after that novel from both 50k and RNN get into Molecular Dynamic Simulation for 5 nanoseconds to filter and 20 nanoseconds toward the novel compounds, then the author deploys all of this into streamlit app and landing page and provide a detailed model validation in and all links found in GitHub link: https://github.com/phalem/Src .