Bayesian compositional regression with microbiome features via variational inference

The microbiome plays a key role in the health of the human body. Interest often lies in finding features of the microbiome, alongside other covariates, which are associated with a phenotype of interest. One important property of microbiome data, which is often overlooked, is its compositionality as it can only provide information about the relative abundance of its constituting components. Typically, these proportions vary by several orders of magnitude in datasets of high dimensions. To address these challenges we develop a Bayesian hierarchical linear log-contrast model which is estimated by mean field Monte-Carlo co-ordinate ascent variational inference (CAVI-MC) and easily scales to high dimensional data. We use novel priors which account for the large differences in scale and constrained parameter space associated with the compositional covariates. A reversible jump Monte Carlo Markov chain guided by the data through univariate approximations of the variational posterior probability of inclusion, with proposal parameters informed by approximating variational densities via auxiliary parameters, is used to estimate intractable marginal expectations. We demonstrate that our proposed Bayesian method performs favourably against existing frequentist state of the art compositional data analysis methods. We then apply the CAVI-MC to the analysis of real data exploring the relationship of the gut microbiome to body mass index.

Socioeconomic determinants of inflammation and neuroendocrine activity: A longitudinal analysis of compositional and contextual effects

Socioeconomic determinants are well-established modulators of inflammation and neuroendocrine activity. Less clear is whether neighbourhood-contextual or individual-compositional factors are more closely associated with gradients in these biomarkers. Here, we examine how immune and neuroendocrine activity are cross-sectionally and longitudinally nested in meso-level socioeconomic characteristics. Participants, male and female, aged ≥50, were recruited from the English Longitudinal Study of Ageing (ELSA). Neighbourhood (Index of Multiple Deprivation [IMD]) and individual (Wealth/Education/Occupational Social Class [Occupation]) factors were drawn from wave 4 (baseline; 2008). Immune and neuroendocrine biomarkers (indexed by C-reactive protein [CRP; n = 3,968]; fibrinogen [n = 3,932]; white blood cell counts [WBCC; n = 4,022]; insulin-like growth factor-1 [IGF-1; n = 4,056]) were measured at baseline and 4-years later (wave 6; 2012). Covariates at baseline included demographic, clinical, and lifestyle variables. Lower socioeconomic status was associated with heighted inflammation and lower neuroendocrine activity unadjusted both cross-sectionally and longitudinally. With few exceptions, cross-sectional associations remained significant after full adjustment. Prospectively, low IMD remained associated with higher CRP and WBCC; wealth with WBCC; and education and occupation with fibrinogen and WBCC. IMD-biomarker associations were reduced when wealth was simultaneously taken into account. Lifestyle accounted for the greatest variance in associations between socioeconomic indicators and inflammation (≤42.11%), but demographics were more salient to neuroendocrine activity (≤88.46%). Neighbourhood-contextual factors were stronger indicators of aberrant biomarker activity than individual-compositional factors in cross-sectional analyses but were largely explained by wealth differences prospectively. Therefore, immune and neuroendocrine changes depended on the composition of the population living in an area, rather than the area itself.

Impact of replacing sedentary behaviour with other movement behaviours on depression and anxiety symptoms: a prospective cohort study in the UK Biobank

BACKGROUND

Sedentary behaviour is potentially a modifiable risk factor for depression and anxiety disorders, but findings have been inconsistent. To assess the associations of sedentary behaviour with depression and anxiety symptoms and estimate the impact of replacing daily time spent in sedentary behaviours with sleep, light, or moderate to vigorous physical activity, using compositional data analysis methods.

METHODS

We conducted a prospective cohort study in 60,235 UK Biobank participants (mean age: 56; 56% female). Exposure was baseline daily movement behaviours (accelerometer-assessed sedentary behaviour and physical activity, and self-reported total sleep). Outcomes were depression and anxiety symptoms (Patient Health Questionnaire-9 and Generalised Anxiety Disorders-7) at follow-up.

RESULTS

Replacing 60 min of sedentary behaviour with light activity, moderate-to-vigorous activity, and sleep was associated with lower depression symptom scores by 1.3% (95% CI, 0.4-2.1%), 12.5% (95% CI, 11.4-13.5%), and 7.6% (95% CI, 6.9-8.4%), and lower odds of possible depression by 0.95 (95% CI, 0.94-0.96), 0.75 (95% CI, 0.74-0.76), and 0.90 (95% CI, 0.90-0.91) at follow-up. Replacing 60 min of sedentary behaviour with moderate-to-vigorous activity and sleep was associated with lower anxiety symptom scores by 6.6% (95% CI, 5.5-7.6%) and 4.5% (95% CI, 3.7-5.2%), and lower odds of meeting the threshold for a possible anxiety disorder by 0.90 (95% CI, 0.89-0.90) and 0.97 (95%CI, 0.96-0.97) at follow-up. However, replacing 60 min of sedentary behaviour with light activity was associated with higher anxiety symptom scores by 4.5% (95% CI, 3.7-5.3%) and higher odds of a possible anxiety disorder by 1.07 (95% CI, 1.06-1.08).

CONCLUSIONS

Sedentary behaviour is a risk factor for increased depression and anxiety symptoms in adults. Replacing sedentary behaviour with moderate-to-vigorous activity may reduce mental health risks, but more work is necessary to clarify the role of light activity.

The difference in sleep, sedentary behaviour, and physical activity between older adults with 'healthy' and 'unhealthy' cardiometabolic profiles: a cross-sectional compositional data analysis approach

Background

Studies have seldom used Compositional Data Analysis (CoDA) to map the effects of sleep, sedentary behaviour, and physical activity on older adults' cardio-metabolic profiles. This study therefore aimed to illustrate how sleep, sedentary behaviour, and physical activity profiles differ between older adult groups (60-89 years), with 'low' compared to those with 'high' concentrations of endocrine cardio-metabolic disease risk markers, using CoDA.

Method

Ninety-three participants (55% female) wore a thigh-mounted triaxial accelerometer for seven consecutive free-living days. Accelerometer estimates of daily average hours of engagement in sedentary behaviour (SB), standing, light-intensity physical activity (LIPA), sporadic moderate-vigorous physical activity (sMVPA, accumulated with bouts between 1 and 10 min), 10-min moderate-vigorous physical activity (₁₀MVPA, accumulated with bouts ≥10 min), in addition to self-reported sleeping hours were reported. Fasted whole blood concentrations of total cholesterol, triglyceride, glucose, and glycated haemoglobin, and serum lipoprotein lipase (LPL), interleukin-6 (IL-6), and procollagen III N-terminal propeptide were determined.

Results

Triglyceride concentration appeared to be highly dependent on ₁₀MVPA engagement as the 'low' and 'high' concentration groups engaged in 48% more and 32% less ₁₀MVPA, respectively, relative to the geometric mean of the entire study sample. Time-use composition of the 'low' LPL group's engagement in ₁₀MVPA was 26% less, while the 'high' LPL group was 7.9% more, than the entire study sample. Time-use composition of the 'high' glucose and glycated haemoglobin groups appeared to be similar as both engaged in more Sleep and SB, and less ₁₀MVPA compared to the study sample. Participants with a 'low' IL-6 concentration engaged in 4.8% more Sleep and 2.7% less ₁₀MVPA than the entire study sample. Time-use composition of the Total Cholesterol groups was mixed with the 'low' concentration group engaging in more Standing and ₁₀MVPA but less Sleep, SB, LIPA, and sMVPA than the entire study sample.

Conclusion

Older adults should aim to increase 10MVPA engagement to improve lipid profile and decrease SB engagement to improve glucose profile.

metaSPARSim: a 16S rRNA gene sequencing count data simulator

BACKGROUND

In the last few years, 16S rRNA gene sequencing (16S rDNA-seq) has seen a surprisingly rapid increase in election rate as a methodology to perform microbial community studies. Despite the considerable popularity of this technique, an exiguous number of specific tools are currently available for proper 16S rDNA-seq count data preprocessing and simulation. Indeed, the great majority of tools have been developed adapting methodologies previously used for bulk RNA-seq data, with poor assessment of their applicability in the metagenomics field. For such tools and the few ones specifically developed for 16S rDNA-seq data, performance assessment is challenging, mainly due to the complex nature of the data and the lack of realistic simulation models. In fact, to the best of our knowledge, no software thought for data simulation are available to directly obtain synthetic 16S rDNA-seq count tables that properly model heavy sparsity and compositionality typical of these data.

RESULTS

In this paper we present metaSPARSim, a sparse count matrix simulator intended for usage in development of 16S rDNA-seq metagenomic data processing pipelines. metaSPARSim implements a new generative process that models the sequencing process with a Multivariate Hypergeometric distribution in order to realistically simulate 16S rDNA-seq count table, resembling real experimental data compositionality and sparsity. It provides ready-to-use count matrices and comes with the possibility to reproduce different pre-coded scenarios and to estimate simulation parameters from real experimental data. The tool is made available at http://sysbiobig.dei.unipd.it/?q=Software#metaSPARSimand https://gitlab.com/sysbiobig/metasparsim.

CONCLUSION

metaSPARSim is able to generate count matrices resembling real 16S rDNA-seq data. The availability of count data simulators is extremely valuable both for methods developers, for which a ground truth for tools validation is needed, and for users who want to assess state of the art analysis tools for choosing the most accurate one. Thus, we believe that metaSPARSim is a valuable tool for researchers involved in developing, testing and using robust and reliable data analysis methods in the context of 16S rRNA gene sequencing.

A critical review of analytical methods in pretreatment of lignocelluloses: Composition, imaging, and crystallinity

Lignocelluloses are widely investigated as renewable substrates to produce biofuels, e.g., ethanol, methane, hydrogen, and butanol, as well as chemicals such as citric acid, lactic acid, and xanthan gum. However, lignocelluloses have a recalcitrance structure to resist microbial and enzymatic attacks; therefore, many physical, thermal, chemical, and biological pretreatment methods have been developed to open up their structure. The efficiency of these pretreatments was studied using a variety of analytical methods that address their image, composition, crystallinity, degree of polymerization, enzyme adsorption/desorption, and accessibility. This paper presents a critical review of the first three categories of these methods as well as their constraints in various applications. The advantages, drawbacks, approaches, practical details, and some points that should be considered in the experimental methods to reach reliable and promising conclusions are also discussed.