Biology subjects

Nishijima, S.

Publications and source records attributed to Nishijima, S..

15 recordsLinked to original sources

bioRxiv2026

Pretraining Objective Shapes Cross-Category Generalization in Affective Image Prediction: A Geometric Comparison of Vision Transformer Encoders

The geometry of representations learned by deep neural networks is shaped jointly by architecture and pretraining objective, yet disentangling these two factors remains difficult. Here we isolate the contribution of pretraining objective by comparing two Vision Transformers from the same backbone family but trained under different objectives: language-image contrastive learning (CLIP) and ImageNet-21k classification. Using continuous Valence-Arousal prediction on the OASIS dataset as a probe of representational quality, we evaluated frozen features under Leave-One-Theme-Out and Leave-One-Category-Out cross-validation, the latter requiring extrapolation to entirely unseen semantic categories. The contrastively pretrained encoder generalized substantially better than the classification-pretrained encoder under both protocols, with the gap widening sharply when held-out categories required cross-category generalization. To characterize why the two representations differ, we developed a geometric analysis of prediction errors, treating per-image errors as vectors in the affective plane and quantifying their spatial structure via weighted phase-locking, trajectory-based occupancy entropy, and effective dimensionality. The classification-pretrained representation collapsed errors into a small number of attractor regions with a strong center-ward pull, whereas the language-aligned representation distributed errors broadly across the affective space. Layer-wise linear probing further revealed that affective information was distributed across depth in the contrastive encoder but increasingly concentrated in deeper layers of the classification encoder, mirroring the texture-bias and category-anchored statistics characteristic of ImageNet-trained representations. These results provide a representation-geometric account of how the choice of pretraining objective, holding architecture constant, determines whether learned features generalize across semantic boundaries or remain confined to category-bound visual regularities. HighlightsO_LIIsolate the effect of pretraining objective by holding the Vision Transformer backbone constant. C_LIO_LIContrastively pretrained features generalize across unseen semantic categories where classification-pretrained features fail. C_LIO_LIIntroduce a geometric analysis of prediction errors based on phase-locking and occupancy entropy. C_LIO_LIClassification pretraining produces concentrated error attractors and a rigid centerward bias. C_LIO_LIAffective information is distributed across depth in CLIP but localized in late layers of the classification ViT. C_LI

neuroscience↗

bioRxiv2026

Global-scale microbiome analyses identify an ancestral gut cyanobacterial lineage enriched in African populations

Industrialization has profoundly reshaped the structure and function of the human gut microbiome, yet our current knowledge of its diversity remains heavily biased toward industrialized populations. Here, we performed an integrated analysis of a large-scale metagenomic dataset (n = 41,471) to systematically characterize the gut microbiome across the African continent. We show that sub-Saharan African populations harbor gut microbiome structures that are distinct not only from industrialized populations but also from other non-industrialized populations. Notably, we identify a strong enrichment of Gastranaerophilales, a poorly characterized lineage of non-photosynthetic Cyanobacteria, in sub-Saharan African populations. Comparative genomic analyses reveal that this lineage encodes distinctive functional features, including motility-related genes, vitamin biosynthesis pathways, and specialized carbohydrate transport systems. Furthermore, we show that this lineage is widely distributed across non-human primates, supporting an evolutionarily conserved host association predating modern humans. Together, our findings identify Gastranaerophilales as a potential ancestral gut symbiont that has been markedly reduced in industrialized societies but retained within African populations, highlighting a major but previously underappreciated component of the human gut microbiome.

microbiology↗

bioRxiv2025

Bent Hockey-Stick Stock-Recruitment Function

The hockey-stick is a simple and widely applicable stock-recruitment function. One drawback is its non-differentiability at the breakpoint, which can be especially problematic when using state-space models for stock assessment. Here, we develop a new bent hockey-stick model that retains the desirable properties of the original hockey-stick function while providing a differentiable breakpoint. We also propose a new algorithm for estimating MSY under multiplicative process stochasticity in the stock-recruitment relationship. Simulation studies demonstrate the utility of these approaches, which can substantially enhance the robustness and sustainability of fisheries management.

ecology↗

bioRxiv2025

Metalog: curated and harmonised contextual data for global metagenomics samples

Metagenomic sequencing enables the in-depth study of microbes and their functions in humans, animals and the environment. While sequencing data is deposited in public databases, the associated contextual data is often not complete and needs to be retrieved from primary publications. This lack of access to sample-level metadata like clinical data or in situ observations impedes cross-study comparisons and meta-analyses. We therefore created the Metalog database, a repository of manually curated metadata for metagenomics samples across the globe. It contains 73,082 samples from humans (including 58,506 of the gut microbiome), 10,703 animal samples, 5,146 ocean water samples, and 21,802 samples from other environmental habitats such as soil, sediment, or fresh water. Samples have been consistently annotated for a set of habitat-specific core features, such as demographics, disease status and medication for humans, host species and captivity status for animals, and filter sizes and salinity for marine samples. Additionally, all original metadata is provided in tabular form, simplifying focused studies e.g. into nutrient concentrations. Pre-computed taxonomic profiles facilitate rapid data exploration, while links to the SPIRE database enable genome-based analyses. The database is freely available for browsing and download at https://metalog.embl.de/.

microbiology↗

bioRxiv2024

Refined Enterotyping Reveals Dysbiosis in Global Fecal Metagenomes

BackgroundEnterotypes describe human fecal microbiomes grouped by similarity into clusters of microbial community composition, often associated with disease, medications, diet, and lifestyle. Numbers and determinants of enterotypes have been derived by diverse frameworks and applied to cohorts that often lack diversity or inter-cohort comparability. ResultsTo overcome these limitations, we selected 16,772 fecal metagenomes collected from 38 countries to revisit the enterotypes using state-of-the-art fuzzy clustering and found robust clustering regardless of underlying taxonomy, consistent with previous findings. Quantifying the strength of enterotype classifications enriched the enterotype landscape, also reflecting some continuity of microbial compositions. As the classification strength was associated with the patients health status, we established an "Enterotype Dysbiosis Score" (EDS) as a latent covariate for various diseases. ConclusionThis global study confirms the enterotypes, reveals a dysbiosis signal within the enterotype landscape, and enables robust classification of metagenomes with an online "Enterotyper" tool, allowing reproducible analysis in future studies. Graphical Abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=79 SRC="FIGDIR/small/607711v3_ufig1.gif" ALT="Figure 1"> View larger version (30K): org.highwire.dtl.DTLVardef@1611a2org.highwire.dtl.DTLVardef@dfbb57org.highwire.dtl.DTLVardef@848ac0org.highwire.dtl.DTLVardef@1b15808_HPS_FORMAT_FIGEXP M_FIG C_FIG

microbiology↗

bioRxiv2024

Fecal microbial load is a major determinant of gut microbiome variation and a confounder for disease associations

The microbiota in individual habitats differ both in relative composition and absolute abundance. While sequencing approaches determine only the relative abundances of taxa and genes, experimental techniques for absolute abundance determination are rarely applied to large-scale microbiome studies. Here, we developed a machine learning approach to predict fecal microbial loads (microbial cells per gram) solely from relative abundance data. Applied to large-scale datasets (n = 34,539), we demonstrate that microbial load is the major determinant of gut microbiome variation and associated with numerous host factors. We found that for several diseases, the altered microbial load, not the disease itself, was the main driver of the gut microbiome changes. Adjusting for this effect substantially reduced the significance of more than half of the disease-associated species. Our analysis reveals that the fecal microbial load is a major confounder in microbiome studies, highlighting its importance for understanding microbiome variation in health and disease.

microbiology↗

bioRxiv2024

Distributional shifts change the biodiversity-ecosystem stability relationship under climate change

Distributional shifts under climate change are increasingly recognized as a biotic change worldwide. However, the effects of distributional shifts on ecosystem variability through changes in biodiversity remain to be clari[fi]ed. In this study, to elucidate the impact of population-level distributional shifts under climate change on community structure and ecosystem function through ecological hierarchy, we investigated the temporal changes in alpha and beta diversity, the contribution of individual species to the observed changes in biodiversity, and the changes in the relationship between biodiversity and ecosystem variability, using over 25 years of data on about 700 species. We found that ecosystem variability was largely stabilized by beta diversity, especially in the present period (2012 to 2021), because beta diversity increased due to the invasion of species that were distributed mainly in tropic and subtropic areas. Despite the effect of beta diversity, ecosystem variability increased in the present period. This increase in ecosystem variability was caused by the spatially synchronized temporal variation in sea-bottom temperature, which resulted in synchronized temporal variation in species abundance among local communities, as well as the increased impact of alpha variability on ecosystem variability. These results indicate that the impact of alpha diversity on ecosystem variability is more changeable than the impact of beta diversity on ecosystem variability under climate change, suggesting the importance of focusing on changes in alpha diversity resulting from local colonization from other habitats and local extinction in existing habitats.

ecology↗

bioRxiv2023

Consistency across multi-omics layers in a drug-perturbed gut microbial community

Multi-omics analyses are increasingly employed in microbiome studies to obtain a holistic view of molecular changes occurring within microbial communities exposed to different conditions. However, it is not always clear to what extent each omics data type contributes to our understanding of the community dynamics and whether they are concordant with each other. Here we map the molecular response of a synthetic community of 32 human gut bacteria to three non-antibiotic drugs by using five omics layers, namely 16S rRNA gene profiling, metagenomics, metatranscriptomics, metaproteomics, and metabolomics. Using this controlled setting, we find that all omics methods with species resolution in their readouts are highly consistent in estimating relative species abundances across conditions. Furthermore, different omics methods complement each other in their ability to capture functional changes in response to the drug perturbations. For example, while nearly all omics data types captured that the antipsychotic drug chlorpromazine selectively inhibits Bacteroidota representatives in the community, the metatranscriptome and metaproteome suggested that the drug induces stress responses related to protein quality control and metabolomics revealed a decrease in polysaccharide uptake, likely caused by Bacteroidota depletion. Taken together, our study provides insights into how multi-omics datasets can be utilised to reveal complex molecular responses to external perturbations in microbial communities.

microbiology↗

bioRxiv2022

Extensive gut virome variation and its associations with host and environmental factors in a population-level cohort

Indigenous bacteriophage communities (virome) in the human gut have a huge impact on the gut bacterial communities (bacteriome), but virome variation at a population scale is not fully investigated yet. Here, we analyse the gut virome in a population-level cohort of 4,198 deeply phenotyped individuals. We discovered thousands of high-quality phage genomes including previously uncharacterized ones with different bacterial hosts than known major phage clades. The distribution of host bacteria was a strong determinant for the distribution of phages, and virome diversity was highly correlated with anti-viral defence mechanisms of the bacteriome, such as CRISPR-Cas and restriction-modification systems. We identified 97 various intrinsic/extrinsic factors that significantly affect the virome, including age, sex, lifestyle, and diet, most of which showed consistent associations with both phages and their predicted bacterial hosts. Among the metadata categories, disease and medication had the strongest effects on the virome structure. Overall, these results present a basis to understand the symbiotic communities of bacteria and their viruses in the human gut, which will facilitate the medical and industrial applications of indigenous viruses.

microbiology↗

bioRxiv2022

Modeling Pulse Dynamics of Juvenile Fish Enables the Short-term Forecasting of Population Dynamics in Japanese Pufferfish: A Latent Variable Approach

The time lag between data collection and management implementation is a source of uncertainty and bias in the calculation of acceptable biological catch. Here, we developed a method to use small-scale survey data for juvenile Japanese pufferfish (Takifugu rubripes) to shorten this time lag and achieve accurate short-term forecasting. A survey of juvenile fish at a local sandy beach in Ise-Mikawa Bay, Japan provides data for the strength of year classes before fisheries recruitment; however, it is difficult to use the raw data owing to the small sample size and large observation errors. We found that a random-effect model overcame these problems and more accurately predicted pulse patterns of catch rates to derive a standardized recruitment index compared with a fixed-effect model. We then showed that a stock assessment model using the standardized recruitment index from the random-effect model outperformed models without the standardized recruitment index with respect to retrospective forecasting ability. This study highlights the applicability of a latent-variable approach for standardizing small-scale survey data and thereby for unbiased forecasting of short-term fish dynamics.

ecology↗

bioRxiv2021

Seasonal variation in the relative importance of assembly processes in marine fish communities as determined by environmental DNA analyses

Compositional variation among local communities reflects the differences in abiotic environments, though the extent of the influence varies in natural systems. Generally, the environmental filtering is expected to be strong if the environmental gradient encompasses severe habitats where species sorting is highly likely to work. However, this hypothesis has rarely been tested in dynamic systems, where the higher dispersal ability of species allows them to easily respond to stressful environments. Here, with the dynamics of fish communities in a Japanese bay revealed by environmental DNA analyses as a model case, we examined how harmful seasonal hypoxia (low concentration of oxygen in bottom waters in summer) affected the strength of environmental filtering. We found that, in summer, dissolved oxygen (DO) concentration was significantly low and fish species richness decreased in the bottom water compared to the surface, suggesting that the organisms were adversely affected by the hypoxia. At the same time, the between-depths heterogeneity in DO concentration was larger, but species composition was less divergent and the influence of DO on species composition appeared weaker during summer. These results imply that environmental filtering is weakened when the bottom water was characterized by extremely severe environments. Furthermore, there was a shift in the species occurrence from bottom to surface waters in summer that was consistent across species, suggesting that the extremely severe hypoxia adversely affected fish species irrespective of their identity. These results collectively suggest that environmental filtering is weaker during summer despite more severity and heterogeneity in environments, most likely because individual movements to avoid unpreferable environments in the bottom waters occurred quasi-neutrally. By providing evidence against the prevailing understanding that environmental filtering strongly works in severe environments, these findings invoke further investigation on how the filtering acts in various conditions.

ecology↗

bioRxiv2020

State-space Modeling Clarifies Productivity Regime Shifts of Japanese Flying Squid

Regime shifts of climatic and environmental conditions potentially affect productivity of fisheries resources, posing challenging issues to stock management. The stocks of the Japanese flying squid (Todarodes pacificus) are suspected to suffer from regime shifts, but their detection is difficult and possibly doubtful because the nature of short-lived species readily makes the effect of regime shifts confounded with observation errors. Here we developed a new state-space assessment model to evaluate the influence of regime shifts on spawner-recruitment relationship of the Japanese flying squid. The model simultaneously estimates the population dynamics of multiple stocks that could share some life history parameters, making parameter inference stable. We demonstrate that two-time regime shifts of productivity around 1991 and 2015 caused two-to three-fold changes of maximum sustinabile yields. The model with regime shifts clarifies the relationship between fishing pressure and spawner abudance that is difficult to detect in a model with no regime shift. The state-space approach will be a promising tool to accurately assess stock status by separating recruitment process from observation errors and contribute tothe management of marine biological resources sensitive to regime shifts.

ecology↗

bioRxiv2020

Spatio-temporal model reduces species misidentification bias of spawning eggs in stock assessment of spotted mackerel in the western North Pacific

Species identification based on morphological characteristics includes species misidentification, leading to estimation bias of population size. The eggs of spotted mackerel Scomber australicus and chub mackerel S. japonicus in the western North Pacific has been identified based on egg diameter. Recent density of spotted mackerel was considerably high despite its low stock biomass. A possibility of this phenomenon is due to overestimation because the difference in egg diameter has become ambiguous between two species. However, we cannot test this possibility using DNA analysis because the eggs are fixed with formalin. Here, we estimated the index of egg density of spotted mackerel using a spatio-temporal model that incorporates the effect of egg density of chub mackerel on the catchability of spotted mackerel, using 15 years data of spawning eggs. We then examined how retrospective biases in estimated stock abundance were reduced when using the index from the model. The index estimated from the model decreased temporal fluctuation and showed smooth patterns. Especially, the recent index was considerably revised down rather than the nominal index. Additionally, the retrospective bias decreased ca. half compared with the nominal index. Therefore, incorporating species misidentification bias should be an essential process for improving stock assessment.

ecology↗

bioRxiv2019

Revealing microbial assemblage structure in the human gut microbiome using latent Dirichlet allocation

Recent research has revealed that there are various microbial species in the human gut microbiome. To clarify the structure of the human gut microbiome, many data mining methods have been applied to microbial composition data. Cluster analysis, one of the key data mining methods that have been used in human gut microbiome research, can classify the human gut microbiome into three clusters, called enterotypes. The human gut microbiome has been suggested to be composed of the microbial assemblages or groups of co-occurring microbes, and one human gut microbiome can contain several microbial assemblages. However, cluster analysis can cluster samples into groups without capturing minor assemblages. In addition, a reliable method of assemblage detection has not been established, and little is known about the distributions of microbial assemblages at a population-level scale. Accordingly, the purpose of this study was to clarify the microbial assemblages in the human gut microbiome. In this study, we detected gut microbiome assemblages using a latent Dirichlet allocation (LDA) method, which was first proposed for the classification of documents in natural language processing. We applied LDA to a large-scale human gut metagenome dataset and found that a four-assemblage LDA model can represent relationships between enterotypes and assemblages with high interpretability. This model indicates that each individual tends to have several assemblages, and each of three assemblages corresponded to each enterotype. However, the C-assemblage can exist in all enterotypes. Interestingly, the dominant genera of the C-assemblage (Clostridium, Eubacterium, Faecalibacterium, Roseburia, Coprococcus, and Butyrivibrio) included butyrate-producing species such as Faecalibacterium prausnitzii. Finally, we revealed that genera mainly appearing in the same assemblage were correlated to each other. We conducted an assemblage analysis on a large-scale human gut metagenome dataset using LDA, a powerful method for detection of microbial assemblages. This approach has the potential to reveal the structure of the human gut microbiome.

microbiology↗

bioRxiv2017

Collaborative environmental DNA sampling from petal surfaces of flowering cherry Cerasus ×- yedoensis 'Somei-yoshino' across the Japanese archipelago

Recent studies have shown that environmental DNA is found almost everywhere. Flower petal surfaces are an attractive tissue to use for investigation of the dispersal of environmental DNA in nature as they are isolated from the external environment until the bud opens and only then can the petal surface accumulate environmental DNA. Here, we performed a crowdsourced experiment, the \"Ohanami Project\", to obtain environmental DNA samples from petal surfaces of Cerasus x yedoensis Somei-yoshino across the Japanese archipelago during spring 2015. C. x yedoensis is the most popular garden cherry species in Japan and clones of this cultivar bloom simultaneously every spring. Data collection spanned almost every prefecture and totaled 577 DNA samples from 149 collaborators. Preliminary amplicon-sequencing analysis showed the rapid attachment of environmental DNA onto the petal surfaces. Notably, we found DNA of other common plant species in samples obtained across a wide distribution; this DNA likely originated from pollen of the Japanese cedar. Our analysis supports our belief that petal surfaces after blossoming are a promising target to reveal the dynamics of environmental DNA in nature. The success of our experiment also shows that crowdsourced environmental DNA analyses have considerable value in ecological studies.

microbiology↗