Postdoctoral Fellow
Campinas, São Paulo, Brasil
• Implemented of an automated pipeline to execute, refine and analyze virtual screening assays that combines molecular dynamics simulations, docking and algorithms which were developed by our group (AutoDock Vina, AmberTools, GROMACS, Python, ArgParse, RdKit, OpenBabel, PyMOL). • Trained deep neural networks with data from molecular dynamics trajectories to build models that predict pharmacokinetic properties of drug candidate molecules (AmberTools, GROMACS, Python, SciKit-Learn, Keras, TensorFlow). • Trained more than 1 million neural network models to optimize its hyperparameters using parallel and distributed computing in clusters (Python, MultiProcessing, Jupyter NoteBooks, MatPloLlib, SeaBorn). • Obtained prediction error (RMSE and MSE) slightly smaller than the best model that was already published in the literature to predict pharmacokinetics properties of molecules that are drug or cosmetic candidates using deep neural networks with the fully connected and convolutional architectures (Python, NumPy, SKLearn, Keras/TensorFlow, RdKit, Mordred). • Mined an online database to build a dataset with more than 100,000 entries about toxicity of molecules (Python, BeautifulSoup4, Pandas).