Metagenomic sequencing projects have produced a glut of sequence data, a large majority (60%-99%) of which is not similar to sequences in public databases. These unknown sequences are referred to as viral dark matter, which are genetic material or proteins of unknown purpose. Traditional bioinformatic techniques are good predictors of function for DNA or protein sequences that are similar to sequences of known function. Unfortunately a majority of metagenome sequences are not similar to sequences in public databases, such as EMBL or GenBank. My research used connectist models to predict the function of unknown sequences or sequences that are not similar to those with functional annotations in public databases. Connectionism is an artificial intelligence approach that models phenomena using networks of interconnected units. The models were inspired by the biological neural networks in the central nervous system of higher organisms. Units typically have weighted connections to other units and each unit modulates its input signal by a transfer function, such as the hyperbolic tangent or the logistic function. Here, I describe my attempts to use two connectist apporaches to prediction the functions of unknown protein sequences. The first model is a feed-forward artificial neural network. The second model is a stacked restricted Boltzmann machine or Deep Network. Both approaches were observed to predict the functions of unknown sequences quite well and some predictions have been validated in the laboratory.