 ProteinFIR is a new method for predicting the function of proteins from their amino acid sequences. It uses deep convolutional neural networks to generate accurate predictions of enzymatic activity, gene ontology terms, and other biological properties. The model can be used as a standalone tool or integrated with existing tools such as BLAST and AlignX. Additionally, the model's ability to generate a generalized functional space makes it ideal for further analysis and interpretation. This article was authored by Theo Sanderson, Maxwell L. Bolesky, David Belanger, and others.