MUNDO: protein function prediction embedded in a multispecies world

Abstract

Motivation Leveraging cross-species information in protein function prediction can add significant power to network-based protein function prediction methods, because so much functional information is conserved across at least close scales of evolution. We introduce MUNDO, a new cross-species co-embedding method that combines a single network embedding method with a co-embedding method to predict functional annotations in a target species, leveraging also functional annotations in a model species network. Results Across a wide range of parameter choices, MUNDO performs best at predicting annotations in the mouse network, when trained on mouse and human PPI networks, in the human network, when trained on human and mouse PPIs, and in Baker’s yeast, when trained on Fission and Baker’s yeast, as compared to competitor methods. MUNDO also outperforms all the cross-species methods when predicting in Fission yeast when trained on Fission and Baker’s yeast; however, in this single case, discarding the information from the other species and using annotations from the Fission yeast network alone usually performs best. Availability All code is publicly available and can be accessed here: github.com/v0rtex20k/MUNDO Supplementary information Supplementary data are available at Bioinformatics Advances online.

Limitations of Current Machine-Learning Models in Predicting Enzymatic Functions for Uncharacterized Proteins

“…Proteins that have similar functions and very similar protein-protein interactions tend to be evolutionarily conserved across species. By creating a shared embedding space that captures the functional and interaction similarities between proteins within and across species, we can transfer functional annotations from well-studied proteins in one species to their functionally similar counterparts in other species 91 . While embedding-based models have shown promise in leveraging multiple data sources for protein function prediction, they have primarily relied on supervised learning approaches that transfer labels from proteins with known functions to unlabeled ones.…”

Joint representation of molecular networks from multiple species improves gene classification

“…Overall, our work demonstrates that we can create effective joint representations of molecular networks from multiple species that can be easily reused for any within-and cross-species gene classification task. We propose GenePlexusZoo as a general framework where any previous [19][20][21][22], or future approaches for joint network embedding and learning can be adapted to take the place of our implementation to continue improving cross-species network-based gene classification.…”

“…Some of these methods directly find similar functional modules in the networks of each species; either through graph alignment [ 17 ] or the use of gene families [ 18 ]. Methods such as MUNK [ 19 ], MUNDO [ 20 ], and ETNA [ 21 ] cast genes across pairs of species ( e . g ., human and mouse) into a joint low dimensional embedding space, anchoring the embedding space together with one-to-one-orthologs.…”

“…Some of these methods directly find similar functional modules in the networks of each species; either through graph alignment [17] or the use of gene families [18]. Methods such as MUNK [19], MUNDO [20], and ETNA [21] cast genes across pairs of species (e.g., human and mouse) into a joint low dimensional embedding space, anchoring the embedding space together with one-to-one-orthologs. A recent deep learning approach, NetQuilt [22], trains a neural network model using a multispecies network representation as input and predicts the full set of gene ontology (GO) biological process [23] terms for each gene.…”