DeepGOZero: Improving protein function prediction from sequence and zero-shot learning based on ontology axioms

Protein functions are often described
  using the Gene Ontology (GO) which is an ontology consisting of over
  50,000 classes and a large set of formal axioms.  Predicting the
  functions of proteins is one of the key challenges in computational
  biology and a variety of machine learning methods have been
  developed for this purpose. However, these methods usually require
  significant amount of training data and cannot make predictions for
  GO classes which
  have only few or no experimental annotations. 
 We developed DeepGOZero, a machine learning model
  which improves predictions for functions with no or only a small
  number of annotations. To achieve this goal, we rely on a
  model-theoretic approach for learning ontology embeddings and
  combine it with neural networks for protein function
  prediction. DeepGOZero can exploit formal axioms in the GO to make
  zero-shot predictions, i.e., predict protein functions even if not a
  single protein in the training phase was associated with that
  function. Furthermore, the zero-shot prediction method employed by
  DeepGOZero is generic and can be applied whenever associations with
  ontology classes need to be predicted.