Protein Function Prediction Through Multi-view Multi-label Latent Tensor Reconstruction

Abstract Background:: While the advent of high-throughput sequencing technologies has accelerated the pace of discovery of proteins, the functions of the majority of them remain unknown. This is mainly due to the time and resource limitations of rigorous experimental functional characterization, which is carried out for only a few select organisms. Consequently, computational methods providing relatively fast and accurate information about the potential cellular activities of a protein are sought after. However, the prediction accuracies of current tools, as assessed through the Critical Assessment of Functional Annotation (CAFA) competition, generally still falls short of the desired level. Results:: We present GO-LTR, a multi-view multi-label prediction model that relies on a high-order tensor approximation of model weights combined with non-linear activation functions. The model is capable of learning high-order relationships between multiple input views representing the proteins and predicting high-dimensional multi-label output consisting of protein functional categories. We demonstrate the competitiveness of our method on various performance measures. Experiments show that GO-LTR learns polynomial combinations between different protein features, resulting in improved performance. Additional investigations establish that GO-LTR shows a high potential in predicting highly specific terms in the gene ontology. Implementation:: The code and data used for training GO-LTR is available at https://github.com/aalto-ics-kepaco/GO-LTR-prediction