Microorganisms often live in complex habitats, where changes in the environment are predictable, providing an opportunity for microorganisms to learn, anticipate the upcoming environmental changes and prepare in advance for better survival and growth. One such environment is the mammalian intestine, where the abundance of different carbon sources is spatially distributed. In this study, we identified seven spatially distributed carbon sources in the mammalian intestine and tested whether Escherichia coli exhibits phenotypes that are consistent with an anticipatory response given their spatial order and abundance within the mammalian intestine. Through RNA-Seq and RT-PCR validation measurements, we found that there was a 67% match in the expression patterns between the measured phenotypes and what would otherwise be expected in the case of anticipatory behavior, while 83% and 0% were in agreement with the homeostatic and random response, respectively. To understand the genetic and phenotypic basis of the discrepancies between the expected and measured anticipatory responses, we thoroughly investigated the discrepancy in D-galactose treatment and the expression of maltose operon in E. coli. Here, the expected anticipatory response, based on the spatial distribution of D-galactose and D-maltose, was that D-galactose should upregulate the maltose operon, but it was the opposite in experimental validation. We performed whole genome random mutagenesis and screening and identified E. coli strains with positive expression of maltose operon in D-galactose. Targeted Sanger sequencing and mutation repair identified that the mutations in the promoter region of malT and in the coding region of the crp gene were the factors responsible for the reversion in the association. Further, to identify why positive association in the D-galactose treatment and the expression of the maltose operon did not evolve naturally, fitness measurements were performed. Fitness experiments demonstrated that the fitness of E. coli strains with a positive association in the D-galactose treatment and the expression of the maltose operon was 12% to 20% lower than that of the wild type strain.

A tantalizing question in cellular physiology is whether the cellular state and environmental conditions can be inferred by the expression signature of an organism. To investigate this relationship, we created an extensive normalized gene expression compendium for the bacterium Escherichia coli that was further enriched with meta-information through an iterative learning procedure. We then constructed an ensemble method to predict environmental and cellular state, including strain, growth phase, medium, oxygen level, antibiotic and carbon source presence. Results show that gene expression is an excellent predictor of environmental structure, with multi-class ensemble models achieving balanced accuracy between 70.0% (±3.5%) to 98.3% (±2.3%) for the various characteristics. Interestingly, this performance can be significantly boosted when environmental and strain characteristics are simultaneously considered, as a composite classifier that captures the inter-dependencies of three characteristics (medium, phase and strain) achieved 10.6% (±1.0%) higher performance than any individual models. Contrary to expectations, only 59% of the top informative genes were also identified as differentially expressed under the respective conditions. Functional analysis of the respective genetic signatures implicates a wide spectrum of Gene Ontology terms and KEGG pathways with condition-specific information content, including iron transport, transferases, and enterobactin synthesis. Further experimental phenotypic-to-genotypic mapping that we conducted for knock-out mutants argues for the information content of top-ranked genes. This work demonstrates the degree at which genome-scale transcriptional information can be predictive of latent, heterogeneous and seemingly disparate phenotypic and environmental characteristics, with far-reaching applications.

A tantalizing question in evolutionary biology is whether evolution can be predicted from past experiences. To address this question, we created a coherent compendium of more than 15,000 mutation events for the bacterium Escherichia coli under 178 distinct environmental settings. Compendium analysis provides a comprehensive view of the explored environments, mutation hotspots and mutation co-occurrence. While the mutations shared across all replicates decrease with the number of replicates, our results argue that the pairwise overlapping ratio remains the same, regardless of the number of replicates. An ensemble of predictors trained on the mutation compendium and tested in forward validation over 35 evolution replicates achieves a 49.2 ± 5.8% (mean ± std) precision and 34.5 ± 5.7% recall in predicting mutation targets. This work demonstrates how integrated datasets can be harnessed to create predictive models of evolution at a gene level and elucidate the effect of evolutionary processes in well-defined environments.

Protein inference, the identification of the protein set that is the origin of a given peptide profile, is a fundamental challenge in proteomics. We present DeepPep, a deep-convolutional neural network framework that predicts the protein set from a proteomics mixture, given the sequence universe of possible proteins and a target peptide profile. In its core, DeepPep quantifies the change in probabilistic score of peptide-spectrum matches in the presence or absence of a specific protein, hence selecting as candidate proteins with the largest impact to the peptide profile. Application of the method across datasets argues for its competitive predictive ability (AUC of 0.80±0.18, AUPR of 0.84±0.28) in inferring proteins without need of peptide detectability on which the most competitive methods rely. We find that the convolutional neural network architecture outperforms the traditional artificial neural network architectures without convolution layers in protein inference. We expect that similar deep learning architectures that allow learning nonlinear patterns can be further extended to problems in metagenome profiling and cell type inference. The source code of DeepPep and the benchmark datasets used in this study are available at https://deeppep.github.io/DeepPep/.

A significant obstacle in training predictive cell models is the lack of integrated data sources. We develop semi-supervised normalization pipelines and perform experimental characterization (growth, transcriptional, proteome) to create Ecomics, a consistent, quality-controlled multi-omics compendium for Escherichia coli with cohesive meta-data information. We then use this resource to train a multi-scale model that integrates four omics layers to predict genome-wide concentrations and growth dynamics. The genetic and environmental ontology reconstructed from the omics data is substantially different and complementary to the genetic and chemical ontologies. The integration of different layers confers an incremental increase in the prediction performance, as does the information about the known gene regulatory and protein-protein interactions. The predictive performance of the model ranges from 0.54 to 0.87 for the various omics layers, which far exceeds various baselines. This work provides an integrative framework of omics-driven predictive modelling that is broadly applicable to guide biological discovery.

Rheumatoid arthritis (RA) is therapeutically challenging due to patient heterogeneity and variability. Herein we describe a novel integration of RA synovial genome-scale transcriptomic profiling of different patient cohorts that can be used to provide predictive insights on drug responses. A normalized compendium consisting of 256 RA synovial samples that cover an intersection of 11,769 genes from 11 datasets was build and compared with similar datasets derived from OA patients and healthy controls. Differentially expression genes (DEGs) that were identified in three independent methods were fed into functional network analysis, with subsequent grouping of the samples based on a non-negative matrix factorization method. RA-relevant pathway activation scores and four machine learning classification techniques supported the generation of a predictive model of patient treatment response. We identified 876 up-regulated DEGs including 24 known genetic risk factors and 8 drug targets. DEG-based subgrouping revealed 3 distinct RA patient clusters with distinct activity signatures for RA-relevant pathways. In the case of infliximab, we constructed a classifier of drug response that was highly accurate with an AUC/AUPR of 0.92/0.86. The most informative pathways in achieving this performance were the NFκB-, FcεRI- TCR-, and TNF signaling pathways. Similarly, the expression of the HMMR, PRPF4B, EVI2A, RAB27A, MALT1, SNX6, and IFIH1 genes contributed in predicting the patient outcome. Construction and analysis of normalized synovial transcriptomic compendia can provide useful insights for understanding RA-related pathway involvement and drug responses for individual patients.

Microbial fermentation is an essential process for research and industrial applications, yet our understanding of cellular dynamics during long-term fermentation is limited. Here, we report a reproducible phenomenon of abrupt population collapse followed by a rapid population rescue that was observed during long-term chemostat cultivations, for various strains of Escherichia coli in minimal media. Through genome resequencing and whole-genome transcriptional profiling of replicate runs over time, we identified that changes in the tRNA and carbon catabolic genes are the genetic basis of this phenomenon. Since current fermentation models are unable to capture the observed dynamics, we present an extended model that takes into account critical biological processes during fermentation, and we further validated carbon source predictions through forward experimentation. This study extends the predictability of current models for microbial fermentation and adds to our system-level knowledge of cellular adaptation during this crucial biotechnological process.

Food and human health are inextricably linked. As such, revolutionary impacts on health have been derived from advances in the production and distribution of food relating to food safety and fortification with micronutrients. During the past two decades, it has become apparent that the human microbiome has the potential to modulate health, including in ways that may be related to diet and the composition of specific foods. Despite the excitement and potential surrounding this area, the complexity of the gut microbiome, the chemical composition of food, and their interplay in situ remains a daunting task to fully understand. However, recent advances in high-throughput sequencing, metabolomics profiling, compositional analysis of food, and the emergence of electronic health records provide new sources of data that can contribute to addressing this challenge. Computational science will play an essential role in this effort as it will provide the foundation to integrate these data layers and derive insights capable of revealing and understanding the complex interactions between diet, gut microbiome, and health. Here, we review the current knowledge on diet-health-gut microbiota, relevant data sources, bioinformatics tools, machine learning capabilities, as well as the intellectual property and legislative regulatory landscape. We provide guidance on employing machine learning and data analytics, identify gaps in current methods, and describe new scenarios to be unlocked in the next few years in the context of current knowledge.

Background

Advanced machine learning methods combined with large sets of health screening data provide opportunities for diagnostic value in human and veterinary medicine.

Hypothesis/objectives

To derive a model to predict the risk of cats developing chronic kidney disease (CKD) using data from electronic health records (EHRs) collected during routine veterinary practice.

Animals

A total of 106 251 cats that attended Banfield Pet Hospitals between January 1, 1995, and December 31, 2017.

Methods

Longitudinal EHRs from Banfield Pet Hospitals were extracted and randomly split into 2 parts. The first 67% of the data were used to build a prediction model, which included feature selection and identification of the optimal neural network type and architecture. The remaining unseen EHRs were used to evaluate the model performance.

Results

The final model was a recurrent neural network (RNN) with 4 features (creatinine, blood urea nitrogen, urine specific gravity, and age). When predicting CKD near the point of diagnosis, the model displayed a sensitivity of 90.7% and a specificity of 98.9%. Model sensitivity decreased when predicting the risk of CKD with a longer horizon, having 63.0% sensitivity 1 year before diagnosis and 44.2% 2 years before diagnosis, but with specificity remaining around 99%.

Conclusions and clinical importance

The use of models based on machine learning can support veterinary decision making by improving early identification of CKD.