Multi-omics has the promise to provide a detailed molecular picture of biological systems. Although obtaining multi-omics data is relatively easy, methods that analyze such data have been lagging. In this paper, we present an algorithm that uses probabilistic graph representations and external knowledge to perform optimal structure learning and deduce a multifarious interaction network for multi-omics data from a bacterial community. Kefir grain, a microbial community that ferments milk and creates kefir, represents a self-renewing, stable, natural microbial community. Kefir has been shown to have a wide range of health benefits. We obtained a controlled bacterial community using the two most abundant and well-studied species in kefir grains: Lentilactobacillus kefiri and Lactobacillus kefiranofaciens. We applied growth temperatures of 30 °C and 37 °C and obtained transcriptomic, metabolomic, and proteomic data for the same 20 samples (10 samples per temperature). We obtained a multi-omics interaction network, which generated insights that would not have been possible with single-omics analysis. We identified interactions among transcripts, proteins, and metabolites, suggesting active toxin/antitoxin systems. We also observed multifarious interactions that involved the shikimate pathway. These observations helped explain bacterial adaptation to different stress conditions, co-aggregation, and increased activation of L. kefiranofaciens at 37 °C.

The diagnosis of inflammatory bowel disease (IBD), ulcerative colitis (UC), and Crohn's disease (CD), relies on clinical and pathological criteria. Non-invasive precision medicine tools to diagnose IBD and discriminate between UC and CD are needed to personalise management. Serum proteomics identified protein biomarkers capable of diagnosing IBD and differentiating CD from UC subtypes. We obtained serum samples from a retrospective study of 47 patients with IBD and non-IBD patients seen in a tertiary care paediatric gastroenterology clinic and applied SomaScan proteomics to measure 1305 proteins to discriminate between IBD and non-IBD and UC and CD. Four proteins were further validated by immunoassays in two retrospective cohorts of 295 and 105 individuals and multi-protein predictors were developed using Support Vector Machines (SVM). The SomaScan discovery phase identified 95 serum protein biomarkers (BH p < 0.01) that differentiated IBD from non-IBD and 70 proteins (p < 0.01) that distinguished UC from CD. Pathway analysis linked specific inflammatory processes and vascular functions to IBD and UC versus CD. An 8-protein classifier achieved an AUC of 0.95 for identifying IBD. Significant elevation of four key predictor proteins (MMP1, MMP3, Resistin, Haptoglobin) in IBD was validated by ELISA in the expanded cohort (N = 295). The 4-protein SVM predictor achieved an AUC of 0.86 and 0.90 for IBD discrimination in two independent cohorts. A separate 4-protein SVM predictor for differentiating UC from CD achieved an AUC of 0.93 in independent validation. Patients with paediatric-onset IBD have a unique serum protein signature associated with pro-inflammatory and vascular pathways. Additional studies are needed to determine whether these dysregulated proteins can be used in conjunction with traditional risk factors to support non-invasive biomarkers that identify IBD and discriminate between its subtypes. Martin Schlaff, The Diane and Dorothy Brooks Foundation, The Manessis Family, Thomas and Lynn Kuzma, and The Hasso Serrano Foundation.