Broad functional profiling of fission yeast proteins using phenomics and machine learning

Reviewer #1 (Public Review):

In this manuscript, the authors aimed to provide information about the likely function of uncharacterised genes in fission yeast. The authors highlight the bias in the literature to well-studied genes/proteins and the fact that the functions of many proteins that are conserved from yeast to humans remain unknown. Initial functional characterisation could provide the impetus for researchers to dedicate time and resources to detailed investigations of protein function. The authors subject the fission yeast deletion set to a battery of perturbations (drug treatments etc) and measured the resultant colony size. In total, 131 conditions were analysed for nearly 3,500 mutants, representing a rich dataset. Clustering analysis was then used to identify common phenotype patterns and thereby infer protein functions using a "guilt by association approach. To assign potential GO terms to uncharacterised proteins, the authors developed a new computational approach (NET-FF) which combined two previous approaches, which they validated against curated annotations on the S. pombe database Pombase. Finally, the authors chose a group of genes which their analysis predicted to be involved in cellular ageing for experimental validation, cross-validating a priority unstudied novel gene (SPAC23C4.09c) to be involved in this process. Overall, the functional analysis performed in this manuscript is rigorous, thorough and incorporates some novel approaches leading to new insights and predicted protein functions. It will be an important resource for the fission yeast community.

Reviewer #2 (Public Review):

This manuscript describes colony-growth phenotypes to measure the fitness of deletion mutants for 3509 non-essential S. pombe genes in 131 conditions. 3492 mutants, including 124 mutants of 'priority unstudied' proteins conserved in humans, providing varied functional clues.

Phenotype-correlation networks provide evidence for the roles of poorly characterized proteins through guilt by association with known proteins. Gene Ontology (GO) terms were predicted using machine learning methods that take advantage of protein-network and protein-homology data.

Integrated analyses produced 1,675 novel GO predictions for 783 genes, including 47 predictions for 23 priority unstudied proteins. Experimental validation for genes involved in cellular ageing were obtained.

A method called NET-FF, which combines network embeddings and protein homology data to predict GO annotations, was developed. The authors demonstrate NET-FF predicts GO terms better than random and compare the information content of the predicted terms with the PomBase GO annotations. The phenotypic data was used to filter the GO annotation predictions made by NET-FF and then explore specific biological examples supported by both datasets

This is a very impressive and rich resource of phenotypic data and it will be particularly useful for the S. pombe research community and generally useful for the functional characterization of highly conserved eukaryotic genes. Overall, the analysis is powerful and sound.

Reviewer #3 (Public Review):

Fission yeast is an important model organism and studies on fission yeast have provided many key insights into the understanding of genes and biological pathways. However, even in such a well-studied model organism, there are still many genes without known functions.

In this work, the authors took advantage of the availability of genome-wide fission yeast deletion mutants to systematically analyze the mutant phenotypes under 131 different conditions. This effort generated a genotype-phenotype dataset larger than the currently curated genotype-phenotype dataset, which is derived from studies over many decades by hundreds of fission yeast laboratories. The authors used the dataset to construct gene clusters that provide functional clues for many genes without previously known functions, including ones conserved in humans. This rich resource will surely be highly useful to the fission yeast community and beyond.

In addition, the authors also used machine learning to generate functional predictions of fission yeast genes and yield novel understandings, which are validated by experimental analysis of new ageing-related genes.

Overall, this study provides unprecedented and highly valuable resources for understanding fission yeast gene functions.