Machine learning on the road to unlocking microbiota's potential for boosting immune checkpoint therapy.

The intestinal microbiota is a complex and diverse ecological community that fulfills multiple functions and substantially impacts human health. Despite its plasticity, unfavorable conditions can cause perturbations leading to so-called dysbiosis, which have been connected to multiple diseases. Unfortunately, understanding the mechanisms underlying the crosstalk between those microorganisms and their host is proving to be difficult. Traditionally used bioinformatic tools have difficulties to fully exploit big data generated for this purpose by modern high throughput screens. Machine Learning (ML) may be a potential means of solving such problems, but it requires diligent application to allow for drawing valid conclusions. This is especially crucial as gaining insight into the mechanistic basis of microbial impact on human health is highly anticipated in numerous fields of study. This includes oncology, where growing amounts of studies implicate the gut ecosystems in both cancerogenesis and antineoplastic treatment outcomes. Based on these reports and first signs of clinical benefits related to microbiota modulation in human trials, hopes are rising for the development of microbiome-derived diagnostics and therapeutics. In this mini-review, we're inspecting analytical approaches used to uncover the role of gut microbiome in immune checkpoint therapy (ICT) with the use of shotgun metagenomic sequencing (SMS) data.

A faecal microbiota signature with high specificity for pancreatic cancer.

BACKGROUND: Recent evidence suggests a role for the microbiome in pancreatic ductal adenocarcinoma (PDAC) aetiology and progression. OBJECTIVE: To explore the faecal and salivary microbiota as potential diagnostic biomarkers. METHODS: We applied shotgun metagenomic and 16S rRNA amplicon sequencing to samples from a Spanish case-control study (n=136), including 57 cases, 50 controls, and 29 patients with chronic pancreatitis in the discovery phase, and from a German case-control study (n=76), in the validation phase. RESULTS: Faecal metagenomic classifiers performed much better than saliva-based classifiers and identified patients with PDAC with an accuracy of up to 0.84 area under the receiver operating characteristic curve (AUROC) based on a set of 27 microbial species, with consistent accuracy across early and late disease stages. Performance further improved to up to 0.94 AUROC when we combined our microbiome-based predictions with serum levels of carbohydrate antigen (CA) 19-9, the only current non-invasive, Food and Drug Administration approved, low specificity PDAC diagnostic biomarker. Furthermore, a microbiota-based classification model confined to PDAC-enriched species was highly disease-specific when validated against 25 publicly available metagenomic study populations for various health conditions (n=5792). Both microbiome-based models had a high prediction accuracy on a German validation population (n=76). Several faecal PDAC marker species were detectable in pancreatic tumour and non-tumour tissue using 16S rRNA sequencing and fluorescence in situ hybridisation. CONCLUSION: Taken together, our results indicate that non-invasive, robust and specific faecal microbiota-based screening for the early detection of PDAC is feasible.

Advanced Dental Cleaning is Associated with Reduced Risk of COPD Exacerbations - A Randomized Controlled Trial.

PURPOSE: Infections from the oral microbiome may lead to exacerbations of chronic obstructive pulmonary disease (COPD). We investigated whether advanced dental cleaning could reduce exacerbation frequency. Secondary outcomes were disease-specific health status, lung function, and whether the bacterial load and composition of plaque microbiome at baseline were associated with a difference in outcomes. PATIENTS AND METHODS: One-hundred-one primary and secondary care patients with COPD were randomized to intervention with advanced dental cleaning or to dental examination only, repeated after six months. At baseline and at 12 months, data of exacerbations, lung function, COPD Assessment Test (CAT) score, and periodontal status were collected from questionnaires, record review, and periodontal examination. Student's t-test and Mann-Whitney-U (MWU) test compared changes in outcomes. The primary outcome variable was also assessed using multivariable linear regression with adjustment for potential confounders. Microbiome analyses of plaque samples taken at baseline were performed using Wilcoxon signed ranks tests for calculation of alpha diversity, per mutational multivariate analysis of variance for beta diversity, and receiver operating characteristic curves for prediction of outcomes based on machine learning models. RESULTS: In the MWU test, the annual exacerbation frequency was significantly reduced in patients previously experiencing frequent exacerbations (p = 0.020) and in those with repeated advanced dental cleaning (p = 0.039) compared with the non-treated control group, but not in the total population including both patients with a single and repeated visits (p = 0.207). The result was confirmed in multivariable linear regression, where the risk of new exacerbations was significantly lower in patients both in the intention to treat analysis (regression coefficient 0.36 (95% CI 0.25-0.52), p < 0.0001) and in the population with repeated dental cleaning (0.16 (0.10-0.27), p < 0.0001). The composition of microbiome at baseline was moderately predictive of an increased risk of worsened health status at 12 months (AUC = 0.723). CONCLUSION: Advanced dental cleaning is associated with a reduced frequency of COPD exacerbations. Regular periodontal examination and dental cleaning may be of clinical importance to prevent COPD exacerbations.

Microbiome meta-analysis and cross-disease comparison enabled by the SIAMCAT machine learning toolbox.

The human microbiome is increasingly mined for diagnostic and therapeutic biomarkers using machine learning (ML). However, metagenomics-specific software is scarce, and overoptimistic evaluation and limited cross-study generalization are prevailing issues. To address these, we developed SIAMCAT, a versatile R toolbox for ML-based comparative metagenomics. We demonstrate its capabilities in a meta-analysis of fecal metagenomic studies (10,803 samples). When naively transferred across studies, ML models lost accuracy and disease specificity, which could however be resolved by a novel training set augmentation strategy. This reveals some biomarkers to be disease-specific, with others shared across multiple conditions. SIAMCAT is freely available from siamcat.embl.de .

Meta-analysis of fecal metagenomes reveals global microbial signatures that are specific for colorectal cancer.

Association studies have linked microbiome alterations with many human diseases. However, they have not always reported consistent results, thereby necessitating cross-study comparisons. Here, a meta-analysis of eight geographically and technically diverse fecal shotgun metagenomic studies of colorectal cancer (CRC, n = 768), which was controlled for several confounders, identified a core set of 29 species significantly enriched in CRC metagenomes (false discovery rate (FDR) < 1 × 10-5). CRC signatures derived from single studies maintained their accuracy in other studies. By training on multiple studies, we improved detection accuracy and disease specificity for CRC. Functional analysis of CRC metagenomes revealed enriched protein and mucin catabolism genes and depleted carbohydrate degradation genes. Moreover, we inferred elevated production of secondary bile acids from CRC metagenomes, suggesting a metabolic link between cancer-associated gut microbes and a fat- and meat-rich diet. Through extensive validations, this meta-analysis firmly establishes globally generalizable, predictive taxonomic and functional microbiome CRC signatures as a basis for future diagnostics.

FitTetra 2.0 - improved genotype calling for tetraploids with multiple population and parental data support.

BACKGROUND: Genetic studies in tetraploids are lagging behind in comparison with studies of diploids as the complex genetics of tetraploids require much more elaborated computational methodologies. Recent advancements in development of molecular techniques and computational tools facilitate new methods for automated, high-throughput genotype calling in tetraploid species. We report on the upgrade of the widely-used fitTetra software aiming to improve its accuracy, which to date is hampered by technical artefacts in the data. RESULTS: Our upgrade of the fitTetra package is designed for a more accurate modelling of complex collections of samples. The package fits a mixture model where some parameters of the model are estimated separately for each sub-collection. When a full-sib family is analyzed, we use parental genotypes to predict the expected segregation in terms of allele dosages in the offspring. More accurate modelling and use of parental data increases the accuracy of dosage calling. We tested the package on data obtained with an Affymetrix Axiom 60 k array and compared its performance with the original version and the recently published ClusterCall tool, showing that at least 20% more SNPs could be called with our updated. CONCLUSION: Our updated software package shows clearly improved performance in genotype calling accuracy. Estimation of mixing proportions of the underlying dosage distributions is separated for full-sib families (where mixture proportions can be estimated from the parental dosages and inheritance model) and unstructured populations (where they are based on the assumption of Hardy-Weinberg equilibrium). Additionally, as the distributions of signal ratios of the dosage classes can be assumed to be the same for all populations, including parental data for some subpopulations helps to improve fitting other populations as well. The R package fitTetra 2.0 is freely available under the GNU Public License as Additional file with this article.

Computational identification of co-evolving multi-gene modules in microbial biosynthetic gene clusters.

The biosynthetic machinery responsible for the production of bacterial specialised metabolites is encoded by physically clustered group of genes called biosynthetic gene clusters (BGCs). The experimental characterisation of numerous BGCs has led to the elucidation of subclusters of genes within BGCs, jointly responsible for the same biosynthetic function in different genetic contexts. We developed an unsupervised statistical method able to successfully detect a large number of modules (putative functional subclusters) within an extensive set of predicted BGCs in a systematic and automated manner. Multiple already known subclusters were confirmed by our method, proving its efficiency and sensitivity. In addition, the resulting large collection of newly defined modules provides new insights into the prevalence and putative biosynthetic role of these modular genetic entities. The automated and unbiased identification of hundreds of co-evolving group of genes is an essential breakthrough for the discovery and biosynthetic engineering of high-value compounds.

reGenotyper: Detecting mislabeled samples in genetic data.

In high-throughput molecular profiling studies, genotype labels can be wrongly assigned at various experimental steps; the resulting mislabeled samples seriously reduce the power to detect the genetic basis of phenotypic variation. We have developed an approach to detect potential mislabeling, recover the "ideal" genotype and identify "best-matched" labels for mislabeled samples. On average, we identified 4% of samples as mislabeled in eight published datasets, highlighting the necessity of applying a "data cleaning" step before standard data analysis.

Pheno2Geno - High-throughput generation of genetic markers and maps from molecular phenotypes for crosses between inbred strains.

BACKGROUND: Genetic markers and maps are instrumental in quantitative trait locus (QTL) mapping in segregating populations. The resolution of QTL localization depends on the number of informative recombinations in the population and how well they are tagged by markers. Larger populations and denser marker maps are better for detecting and locating QTLs. Marker maps that are initially too sparse can be saturated or derived de novo from high-throughput omics data, (e.g. gene expression, protein or metabolite abundance). If these molecular phenotypes are affected by genetic variation due to a major QTL they will show a clear multimodal distribution. Using this information, phenotypes can be converted into genetic markers. RESULTS: The Pheno2Geno tool uses mixture modeling to select phenotypes and transform them into genetic markers suitable for construction and/or saturation of a genetic map. Pheno2Geno excludes candidate genetic markers that show evidence for multiple possibly epistatically interacting QTL and/or interaction with the environment, in order to provide a set of robust markers for follow-up QTL mapping. We demonstrate the use of Pheno2Geno on gene expression data of 370,000 probes in 148 A. thaliana recombinant inbred lines. Pheno2Geno is able to saturate the existing genetic map, decreasing the average distance between markers from 7.1 cM to 0.89 cM, close to the theoretical limit of 0.68 cM (with 148 individuals we expect a recombination every 100/148=0.68 cM); this pinpointed almost all of the informative recombinations in the population. CONCLUSION: The Pheno2Geno package makes use of genome-wide molecular profiling and provides a tool for high-throughput de novo map construction and saturation of existing genetic maps. Processing of the showcase dataset takes less than 30 minutes on an average desktop PC. Pheno2Geno improves QTL mapping results at no additional laboratory cost and with minimum computational effort. Its results are formatted for direct use in R/qtl, the leading R package for QTL studies. Pheno2Geno is freely available on CRAN under "GNU GPL v3". The Pheno2Geno package as well as the tutorial can also be found at: http://pheno2geno.nl .

WormQTLHD--a web database for linking human disease to natural variation data in C. elegans.

Interactions between proteins are highly conserved across species. As a result, the molecular basis of multiple diseases affecting humans can be studied in model organisms that offer many alternative experimental opportunities. One such organism-Caenorhabditis elegans-has been used to produce much molecular quantitative genetics and systems biology data over the past decade. We present WormQTL(HD) (Human Disease), a database that quantitatively and systematically links expression Quantitative Trait Loci (eQTL) findings in C. elegans to gene-disease associations in man. WormQTL(HD), available online at http://www.wormqtl-hd.org, is a user-friendly set of tools to reveal functionally coherent, evolutionary conserved gene networks. These can be used to predict novel gene-to-gene associations and the functions of genes underlying the disease of interest. We created a new database that links C. elegans eQTL data sets to human diseases (34 337 gene-disease associations from OMIM, DGA, GWAS Central and NHGRI GWAS Catalogue) based on overlapping sets of orthologous genes associated to phenotypes in these two species. We utilized QTL results, high-throughput molecular phenotypes, classical phenotypes and genotype data covering different developmental stages and environments from WormQTL database. All software is available as open source, built on MOLGENIS and xQTL workbench.

Mechanism of selective halogenation by SyrB2: a computational study.

The mechanism of the chlorination reaction of SyrB2, a representative α-ketoglutarate dependent halogenase, was studied with computational methods. First, a macromolecular model of the Michaelis complex was constructed using molecular docking procedures. Based on this structure, a smaller model comprising the first- and some of the second-shell residues of iron and a model substrate was constructed and used in DFT investigations on the reaction mechanism. Computed relative energies and Mössbauer isomer shifts as well as quadrupole splittings indicate that the two oxoferryl species observed experimentally are two stereoisomers resulting from an exchange of the coordination sites occupied by the oxo and chloro ligands. In principle both Fe(IV)═O species are reactive and decay to Fe(III)Cl (OH)/carbon radical intermediates via C-H bond cleavage. In the final rebound step, which is very fast and thus precluding equilibration between the two forms of the radical intermediate, the ligand (oxo or chloro) placed closest to the carbon radical (trans to His235) is transferred to the carbon. For the native substrate (L-Thr) the lowest barrier for C-H cleavage was found for an isomer of the oxoferryl species favoring chlorination in the rebound step. CASPT2 calculations for the spin state splittings in the oxoferryl species support the conclusion that once the Fe(IV)═O intermediate is formed, the reaction proceeds on the quintet potential energy surface.