Biology subjects

Koh, H.

Publications and source records attributed to Koh, H..

8 recordsLinked to original sources

bioRxiv2026

Model-Agnostic Decomposition of the Difference in Mean Response Between Two Groups via Game-Theoretic Attribution with Minimum-Switching-Distortion Coupling

The difference in mean response between two groups can arise from differences in covariate-response relationships, covariate distributions, or both. We propose a model-agnostic decomposition (MAD) of this difference into structure and composition effects. We formulate a cooperative switching game with a model-switch player and individual covariate-switch players, organized into model and covariate unions. MAD employs the Owen value, whose quotient game yields the aggregate structure and composition effects and whose internal game allocates the aggregate composition effect across covariates, and thereby inherits its fair allocation properties. In the internal game, partial covariate switches create hybrid covariate vectors whose joint distributions are not determined by the group-specific covariate distributions alone. To resolve this indeterminacy, we propose a minimum-switching-distortion (MSD) coupling, formulated as an optimal-transport problem. The MSD coupling satisfies admissibility conditions of identity at coincidence, transposition equivariance, and coordinate equivariance, and its sparse support reduces allocation cost. Because MAD imposes no particular functional form, it accommodates machine-learning models that adapt to complex nonlinear and interactive relationships, thereby yielding more accurate decompositions. Simulation studies demonstrate the effectiveness of MAD and the MSD coupling in terms of decomposition accuracy, coupling admissibility, and computational cost. As an illustration, we decompose the difference in mean body mass index between UK and US males aged 30-70 years using microbiome data, at the aggregate level and across microbial genera.

bioinformatics↗

bioRxiv2026

Efficient Game-Theoretic Explanations for Tree-Based Ensembles via Owen Values

AO_SCPLOWBSTRACTC_SCPLOWShapley-value-based explanations, notably SHAP (SHapley Additive exPlanations), have gained prominence as a principled game-theoretic framework for local explanations and global feature importance. While exact Shapley value computation is exponential in feature count, TreeExplainer exploits the recursive structure of decision trees to achieve polynomial-time computation for tree-based ensembles. In many scientific applications, however, features are naturally organized into a priori groups reflecting domain knowledge, requiring explanations both across and within groups. The Owen value extends the Shapley value through a two-stage allocation rule that incorporates group structure while preserving fairness properties; yet, efficient algorithms for its computation remain limited. In this paper, we propose exact and Monte Carlo algorithms for computing Owen values in tree-based ensembles by combining hierarchy-guided group aggregation with tree-aware dynamic programming. The exact algorithm computes Owen values without sampling under the path-dependent characteristic function, which approximates the conditional expectation, whereas the Monte Carlo algorithm provides a scalable approximation that is unbiased for any prespecified sampling budget and converges almost surely as the sampling budget increases. We also provide global importance measures and visualization tools for structured, multi-resolution explanations. The proposed algorithms and tools are collectively referred to as TreeOwen. Through simulation experiments, we demonstrate the numerical accuracy and substantial computational gains of TreeOwen. We illustrate its practical utility using immunotherapy metagenomic data, showing how microbial genera (groups) and species (features) contribute to patient recovery.

bioinformatics↗

bioRxiv2026

Cross-cohort analysis reveals conserved gut virome signatures and phage-encoded auxiliary functions in ulcerative colitis

While gut bacteriome dysbiosis is a well-established hallmark of ulcerative colitis (UC), the ecological and functional remodeling of the gut virome conserved across diverse populations remains unclear. Here, by constructing a cross-cohort atlas of the UC-associated fecal virome, we show that viral reorganization closely parallels bacterial dysbiosis, with enrichment of phages targeting disease-associated bacterial taxa. We identified a conserved set of UC-associated viral signatures that robustly distinguished UC from healthy controls using machine-learning-based classification across independent cohorts. Functional profiling further revealed that phages enriched in UC carried higher densities of auxiliary metabolic genes (AMGs) related to virulence and antibiotic resistance than phages depleted in UC, with immune evasion and glycopeptide resistance genes particularly overrepresented. Together, our cross-cohort approach highlights the value of the gut virome-based diagnostic framework for UC and suggests that phage-encoded AMGs may contribute to shaping the gut ecosystem under inflammatory conditions.

microbiology↗

bioRxiv2026

Dual Oxygen-Partitioned Co-Culture Uncovers Microbe-Specific Epithelial Stress and Homeostatic Programs

Direct epithelial-microbial interactions occur across a steep aerobic-anaerobic interface in the intestine, yet mechanistic analysis has been limited by the difficulty of sustaining oxygen-dependent epithelial cells together with strictly anaerobic microbes in vitro. Here, we establish a dual oxygen-partitioned co-culture system that reproducibly maintains aerobic intestinal epithelial cells and obligate anaerobic gut bacteria, enabling controlled analysis of epithelial responses under physiologically relevant oxygen architecture. Using Mediterraneibacter gnavus and Lacticaseibacillus casei as contrasting microbial partners, we identify distinct epithelial programs. M. gnavus induces epithelial stress characterized by tight-junction disruption, suppression of mitochondrial respiration, and activation of chromatin-associated and innate immune regulatory pathways, whereas L. casei preserves barrier integrity and supports mitochondrial capacity with restrained immune modulation. Integrated transcriptomic and proteomic analyses reveal a shared epithelial energy-conserving response to microbial proximity, upon which microbe-specific stress or homeostatic programs are imposed. In vivo analysis using a dextran sodium sulfate-induced colitis model demonstrates that these epithelial programs become physiologically relevant under inflammatory conditions, with L. casei promoting epithelial repair and immune rebalancing. Together, these findings define oxygen-partitioned epithelial-microbial interactions as a determinant of microbe-specific epithelial states.

microbiology↗

bioRxiv2025

Early death and neuronal abnormalities in depdc5 loss-of-function mosaic zebrafish models

DEPDC5 (DEP domain-containing protein 5) encodes a repressor of the mTORC1 signaling pathway. Variants in DEPDC5 are associated with a range of focal epilepsies, including mosaic variants associated with focal cortical dysplasia (FCD) and other focal brain malformations with brain-only somatic mosaic variants. To investigate the role of DEPDC5 in human epilepsy related to mosaic variants, we have generated mosaic depdc5 loss-of-function zebrafish models using homology-based constructs acutely targeting depdc5 and labeled with tdTomato to allow for visualization of the degree of mosaicism. The resulting mosaic depdc5 CRISPants demonstrated early larval death, with [~]50% of CRISPants (vs. 10% of controls) dead by 7 days post fertilization (dpf), analogous to the early death sometimes associated with human DEPDC5-related epilepsy. We compared depdc5 CRISPants with uninjected and scrambled controls from the same clutches. Body and head size were reduced in the depdc5 CRISPants. Analysis of swimming behavior showed a striking reduction in distance traveled and maximum velocity in the depdc5 CRISPants vs. controls. Based on visual confirmation of mutational load, we categorized CRISPants into depdc5+ vs. depdc5++, reflecting weak vs. strong tdTomato fluorescence. We observed that depdc5++ CRISPants had increased episodes of posture loss, suggesting increased seizure-like behavior related to higher percentages of mutant cells. Local field potential recordings revealed increased neuronal hyperexcitability in depdc5 CRISPants vs. controls. Acridine orange staining demonstrated early apoptosis in the CRISPants vs. controls. Our mosaic depdc5 CRISPants provide a clinically relevant model to study the role of mosaic DEPDC5-related epilepsy and early death.

neuroscience↗

bioRxiv2025

Phylogeny-Informed Random Forests for Human Microbiome Studies

AO_SCPLOWBSTRACTC_SCPLOWRandom Forest is a widely used tree-based ensemble learning algorithm that efficiently captures complex nonlinear relationships and higher-order feature interactions with no distributional assumptions to be satisfied. It is also well-suited to human microbiome studies, where the data are highly skewed, overdispersed, discrete, and irregular. Here, I pay particular attention to the phylogenetic tree information that reflects evolutionary ancestry and functional relatedness among microbial features. Proper incorporation of phylogenetic tree information into microbiome data analysis has provided new insights and improved analytical performance. In this paper, I introduce an extension of the Random Forest algorithm that incorporates phylogenetic tree information, named Phylogeny-Informed Random Forests (PIRF), to improve predictive accuracy in human microbiome studies. The core mechanism of PIRF lies in its localized approach; rather than treating all features as competing globally to be selected or weighted, PIRF identifies informative features within each phylogenetic cluster (i.e., a localized group of microbial features that are evolutionarily and functionally related), thereby enriching functional representations while reducing tree correlation. I demonstrate the high predictive accuracy of PIRF, compared with other off-the-shelf tools, across seven benchmark tasks: four classification problems (gingival inflammation, immunotherapy response, type 1 diabetes, and obesity) and three regression problems (cytokine level, age based on oral microbiome, and age based on gut microbiome). ImportancePIRF is an extension of the Random Forest algorithm that incorporates phylogenetic tree information to improve predictive accuracy in human microbiome studies. PIRF can serve as a useful tool for microbiome-based disease diagnostics and personalized medicine. The software and tutorials are freely available as an R package, named PIRF, at https://github.com/hk1785/PIRF.

bioinformatics↗

bioRxiv2025

A general kernel machine comparative analysis framework for randomized block designs

MotivationThere are numerous potential confounders, including genetic, environmental, technical, and demographic factors. These factors may be known or unknown, measured or unmeasured; hence, it is extremely challenging to capture them in downstream data analysis. However, randomized block design is an efficient design technique to control confounding factors and to reduce variability within subjects. This helps prevent spurious discoveries and boost test power. I also note that kernel machine comparative analysis is widely employed in high-dimensional omics studies to boost test power by combining possibly weak effects from multiple underlying variants, and also to explore various linear or nonlinear patterns of disparity. ResultsIn this paper, I introduce a general kernel machine comparative analysis framework for randomized block designs, named as KernRBD, to investigate the effects of treatments (e.g., medical treatment, environmental exposure) on the underlying variants. KernRBD is unique in its range of functionalities, including the computation of P-value for global testing and adjusted P-values for pairwise comparisons, as well as visual representation through ordination plotting. KernRBD is practical, requiring only a kernel as input, and also robustly valid based on a resampling scheme not requiring the assumption of normality to be satisfied. I also introduce its omnibus test for a unified and powerful significance testing across multiple input kernels. While its applications should be much broader, I illustrate its use through human microbiome {beta}-diversity analysis in praxis, and its outperformance in significance testing through simulation experiments in silico. Availability and ImplementationKernRBD is available at https://github.com/hk1785/kernrbd.

bioinformatics↗

bioRxiv2025

An ensemble learning method for joint kernel association testing and principal component analysis on multiple kernels

In high-dimensional omics studies, researchers often conduct kernel association testing to power-fully detect the relationship of the genetic or microbial composition with human health or disease. Especially, in human microbiome studies, its dimension reduction analysis follows to visually represent complex microbiome data in a simple two- or three-dimensional coordinate space. However, various kernels exist, and they produce all different outcomes; hence, it is hard to interpret them all consistently. Then, omnibus testing has recently been a subject of intense investigation for a unified and powerful statistical inference. However, current omnibus tests are purely a test for significance producing only a P-value as their outcome with no related dimension reduction and visualization approach; hence, their utility is still limited. In this paper, I introduce an ensemble learning method, named as enKern, for joint kernel associating testing and principal component analysis on multiple kernels. enKern is based on a weight learning scheme that leverages complementary contributions from multiple kernels for powerful performance for various association patterns. I show that applying the weights to individual test statistics or individual kernels is equivalent, which in turn enables a visualization in a reduced dimensional coordinate space based on the weighted kernel to be matched with its original significance testing scheme. I demonstrate its use for human microbiome {beta}-diversity analysis. I also demonstrate its outperformance in validity and power through simulation experiments. enKern is freely available in R computing environment at https://github.com/hk1785/enkern.

bioinformatics↗