Author: Wen Torng and Russ B. Altman, Stanford University 2018 We provide datasets and implementation for reproducing experiments descirbed in "High Precision Protein Functional Site Detection using 3D Convolutional Neural Network" paper. The models are built upon Theano version '0.8.2' The code are organized into four different modules corresponding to three main experiments performed in the paper, and code for performing user defined tests using the trained models. (1) PROSITE: Models for the detecting the 10 PROSITE families (2) Benchmark_GASS: Models for detecting NOS and TRYPSIN_like enzymes (where the TRYPSIN_SER.6 and TRYPSIN_HIS.5 models are detectly copied from the PROSITE module) (3) Detect_ATP_sites: Models for detecting ATP binding residues (4) User_test: Interface for performing detection on structures provided by users using the trained modeuls from the above modules. Under each module, code for for dataset generation, model training, model evalutation and the generated results are provided. Documentation for each module can be found in README files under each module.