Biology subjects

Tikhonov, G.

Publications and source records attributed to Tikhonov, G..

5 recordsLinked to original sources

bioRxiv2024

Common to rare transfer learning (CORAL) enables inference and prediction for a quarter million rare Malagasy arthropods

Modern DNA-based biodiversity surveys result in massive-scale data, including up to millions of species - of which most are rare. Making the most of such data for inference and prediction requires modelling approaches that can relate species occurrences to environmental and spatial predictors, while incorporating information about their taxonomic or phylogenetic placement. Even if the scalability of joint species distribution models to large communities has greatly advanced, incorporating hundreds of thousands of species has not been feasible to date, leading to compromised analyses. Here we present a novel "common to rare transfer learning" approach (CORAL), based on borrowing information from the common species to enable statistically and computationally efficient modelling of both common and rare species. We illustrate that CORAL leads to much improved prediction and inference in the context of DNA metabarcoding data from Madagascar, comprising 255,188 arthropod species detected in 2874 samples.

ecology↗

bioRxiv2024

Accelerating joint species distribution modeling with Hmsc-HPC: A 1000x faster GPU deployment

Joint Species Distribution Modelling (JSDM) is a powerful and increasingly widely used statistical methodology in biodiversity modelling, enabling researchers to assess and predict the joint distribution of species across space and time. However, JSDM can be computationally intensive and even prohibitive, especially for large datasets and sophisticated model structures. To address computational limitations of JSDM, we expanded one widely used JSDM framework, Hmsc-R, by developing a Graphical Processing Unit (GPU) -compatible implementation of its model fitting algorithm. While our augmented framework retains the original user interface in R, its new computational core is coded in Python and dominantly uses TensorFlow library. This enhancement primarily targets to enable leveraging high-performance computing resources effectively, though it also accelerates model fitting with consumer-level machines. This upgrade is designed to leverage high-performance computing resources more effectively. We evaluated the performance of the proposed implementation across diverse model configurations and dataset sizes. Our results indicate significant model fitting speed-up compared to the existing Hmsc-R package across most models. Notably, for the largest datasets, we achieved >1000 times speed-ups. This GPU-compatible enhancement boosts the scalability of Hmsc-R package by several orders of magnitude, reaching a significantly higher level. It opens promising opportunities for modeling extensive and intricate datasets, enabling better-informed conservation strategies, environmental management, and climate change adaptation planning. Author summaryOur study addresses the computational challenges associated with Joint Species Distribution Modelling (JSDM), a critical statistical methodology for understanding species distributions in biodiversity research. Despite its utility, JSDM often faces computational limitations, particularly for large datasets. To overcome this hurdle, we enhance the widely used Hmsc-R framework by introducing a GPU-compatible implementation of its model fitting algorithm. Our upgraded framework, while retaining the user-friendly R interface, leverages Python and TensorFlow for its computational core, enabling efficient utilization of high-performance computing resources. Through extensive evaluation across diverse model configurations and dataset sizes, we demonstrate substantial speed-ups compared to the original Hmsc-R package, with over 1000 times speed-ups observed for the largest datasets. This GPU-compatible enhancement significantly improves the scalability of JSDM, enabling the analysis of extensive and complex biodiversity datasets. Our work has far-reaching implications for informing conservation strategies, environmental management, and climate change adaptation planning by facilitating more efficient and accurate biodiversity modeling, ultimately contributing to better-informed decision-making in ecological research and practice.

ecology↗

bioRxiv2021

Scalable estimation of microbial co-occurrence networks with Variational Autoencoders

Estimating microbe-microbe interactions is critical for understanding the ecological laws governing microbial communities. Rapidly decreasing sequencing costs have promised new opportunities to estimate microbe-microbe interactions across thousands of uncultured, unknown microbes. However, typical microbiome datasets are very high dimensional and accurate estimation of microbial correlations requires tens of thousands of samples, exceeding the computational capabilities of existing methodologies. Furthermore, the vast majority of microbiome studies collect compositional metagenomics data which enforces a negative bias when computing microbe-microbe correlations. The Multinomial Logistic Normal (MLN) distribution has been shown to be effective at inferring microbe-microbe correlations, however scalable Bayesian inference of these distributions has remained elusive. Here, we show that carefully constructed Variational Autoencoders (VAEs) augmented with the Isometric Log-ratio (ILR) transform can estimate low-rank MLN distributions thousands of times faster than existing methods. These VAEs can be trained on tens of thousands of samples, enabling co-occurrence inference across tens of thousands of microbes without regularization. The latent embedding distances computed from these VAEs are competitive with existing beta-diversity methods across a variety of mouse and human microbiome classification and regression tasks, with notable improvements on longitudinal studies.

bioinformatics↗

bioRxiv2019

Joint species distribution modelling with HMSC-R

O_LIJoint Species Distribution Modelling (JSDM) is becoming an increasingly popular statistical method for analyzing data in community ecology. JSDM allow the integration of community ecology data with data on environmental covariates, species traits, phylogenetic relationships, and the spatio-temporal context of the study, providing predictive insights into community assembly processes from non-manipulative observational data of species communities. Hierarchical Modelling of Species Communities (HMSC) is a general and flexible framework for fitting JSDMs, yet its full range of functionality has remained restricted to Matlab users only.\nC_LIO_LITo make HMSC accessible to the wider community of ecologists, we introduce HMSC-R 3.0, a user-friendly R implementation of the framework described in Ovaskainen et al (Ecology Letters, 20 (5), 561-576, 2017) and further extended in several later publications.\nC_LIO_LIWe illustrate the use of the package by providing a series of five vignettes that apply HMSC-R 3.0 to simulated and real data. HMSC-R applications to simulated data involve single-species models, models of small communities, and models of large species communities. They demonstrate the estimation of species responses to environmental covariates and how these depend on species traits, as well as the estimation of residual species associations. They further demonstrate how HMSC-R can be applied to normally distributed data, count data, and presence-absence data. The real data consist of bird counts in a spatio-temporally structured dataset, environmental covariates, species traits and phylogenetic relationships. The vignettes demonstrate how to construct and fit many kinds of models, how to examine MCMC convergence, how to examine the explanatory and predictive powers of the models, how to assess parameter estimates, and how to make predictions.\nC_LIO_LIThe package, along with the extended vignettes, makes JSDM fitting and post-processing easily accessible to ecologists familiar with R.\nC_LI

ecology↗

bioRxiv2019

Variation in Melitaea cinxia gut microbiota is phylogenetically highly structured but only mildly driven by host plant microbiota, sex or parasitism.

Originality-Significance StatementThe factors contributing to the assembly of microbiota in animals are extremely complex, and thus a comprehensive understanding of the mechanisms shaping host-associated microbial communities in natural ecosystems requires extensive ecological studies and appropriate statistical methods. In this study, we investigated the bacterial microbiota associated with the caterpillars of the Glanville fritillary (Melitaea cinxia), which is a long-term-studied ecological model system. We assessed the structure of variation in both occurrence and abundance of gut microbial communities of individuals collected in the wild with joint-species modelling, with the aim to relate the microbial community structure with multiple potentially impacting covariates: host plant microbiota and metabolites, hosts sex, potential parasitoid infection, and family structure. These covariates exhibited substantial correlation with multiple microbial taxas occurrences, which correlations were consistent for phylogenetically related groups of taxa, but varied across the whole microbial community; on the contrary, only few correlations were found with taxas abundances. The dominating co-occurrence pattern of microbiota assembly, which effectively split caterpillar individuals into two distinct groups, was, however, unrelated to any of the considered covariates.\n\nSummaryUnderstanding of what ecological factors shape intraspecific variation of insect microbiota is still relatively poor. In Lepidopteran caterpillars, microbiota is assumed to be mainly composed of transient bacterial symbionts acquired from the host plant. We sampled Glanville fritillary (Melitaea cinxia) caterpillars from natural populations to describe the microbiome and to identify potential factors that determine the structure of the microbial community, including the sex of the host, the impact of parasitoid infection, and the possible link between host plant and caterpillar microbiota. Our results demonstrate high variability of microbiota composition even among caterpillars that shared the same host plant individual. The observed variation in microbiota composition is partially attributed to the measured properties of the host or its plant microbial and chemical composition, and is aligned with microbial phylogenetic structure, with related taxa exhibiting similar patterns. However, the prevailing part of the observed variation was not associated with any of the assessed characteristics, although it followed a pronounced segregation structure: in some caterpillars the microbial communities were dominated by several related Enterobacteriaceae taxa, while in others these taxa were absent. Our results challenge previous findings that the host plant properties are the major drivers of microbiota communities of insect herbivores.

ecology↗