Genome-scale prediction of gene ontology from mass fingerprints reveals new metabolic gene functions

Mass-based fingerprinting can characterize microorganisms; however, expansion of these methods to predict specific gene functions is lacking. Therefore, mass fingerprinting was developed to functionally profile a yeast knockout library. Matrix-assisted laser desorption/ionization time-of-flight (MALDI-TOF) fingerprints of 3,238 <i>Saccharomyces cerevisiae</i> knockouts were digitized for correlation with gene ontology (GO). Random forests and support vector machine (SVM) algorithms assigned GO terms with average AUC values of 0.994 and 0.980, respectively. SVM was the best predictor with average true-positive and true-negative rates of 0.983 and 0.993, respectively. To test predictions of unknown gene functions, the dataset of uncharacterized yeast gene knockouts was evaluated based on SVM scores, and new functions were suggested for 28 corresponding genes. Metabolomics analysis of two knockouts (YDR215C and YLR122C) of uncharacterized genes predicted to be involved in methylation-related metabolism showed altered intracellular contents of methionine-related metabolites. Increased S-adenosylmethionine in YDR215C indicated that this strain shows potential as a chassis for bioproduction of methylated compounds. This study demonstrates that fingerprinting can generate large functional datasets for improved machine learning-based gene function prediction.