DLKcat is a deep-learning model that predicts enzyme turnover number (kcat) for any enzyme-substrate pair from a protein amino-acid sequence and a substrate SMILES string. It combines a graph neural network (GNN) over the substrate molecular graph with an attention-CNN over the protein sequence, fuses the two representations, and regresses log2(kcat) in s^-1. The original repository (DeeplearningApproach/) ships a trained PyTorch checkpoint plus a single inference script (prediction_for_input.py) that reads a TSV of (substrate name, SMILES, protein sequence) rows and writes one Kcat value per row.