Biology subjects

Lysenko, A.

Publications and source records attributed to Lysenko, A..

15 recordsLinked to original sources

bioRxiv2026

QBayMic: Quantum-coupled variational Bayes for clustering and feature selection in low-signal microbiome data

Clustering microbiome samples into community types is central to cohort stratification and biomarker discovery, yet the resulting inference becomes unstable when the between group signal is small compared with sampling noise: variational Bayes yields different partitions across initialisations, and common fixes do not solve the problem. Simple restarts are ineffective because the variational free energy is anti-correlated with clustering accuracy; deterministic annealing collapses to the same solution as greedy ascent, with the operator staying diagonal at every temperature; and parallel tempering replicas remain too similar to permit configuration exchanges. We propose QBayMic, which replaces the assignment step of a Dirichlet-multinomial mixture with sparse variable selection via a quantum Gibbs state under an annealed Hamiltonian, coupling competing assignments through a transverse-field term that cannot be reproduced by temperature scaling alone. We present two gate-based circuit designs for this step, evaluating on noiseless qubit-register simulations, and we derive a signal fraction, computable prior to clustering, that predicts the expected strength of quantum coupling. With matched compute in the predicted regime, the three classical methods recovered the reference partition (ARI > 0.4) in 0/100 seeds, while QBayMic recovered it in 47-64/100; when the number of clusters exceeded three, only QBayMic recovered the correct cluster count. For a soil pH dataset, the diagnostic indicates a narrow separation margin; for a human-derived dataset tuned into the predicted band via controlled dilution, classical methods recovered the cluster count in 0/100 seeds, compared with 61-76% for QBayMic. The implementation is publicly available at https://github.com/tungtokyo1108/QBayMic.

bioinformatics↗

bioRxiv2026

A hyperspherical deep Bayesian model for interpretable clustering and relationship prediction in microbiome multi-omics integration

The microbiome plays a significant role in the development and progression of many diseases, yet extracting interpretable insights from multi-omics data remains challenging. Existing approaches face a recurring practical trade-off: deep learning methods achieve high predictive performance but lack uncertainty quantification, whereas probabilistic methods provide interpretable results but require data-type-specific likelihood functions that limit generalization across diverse omics modalities. Here, we introduce DBayesCM (Deep Bayesian Clustering for Multi-omics), which combines deep learning modeling with Bayesian nonparametric methods. DBayesCM employs separate encoders to project microbiome and host omics data into a shared latent space, where an infinite mixture model with a Dirichlet process prior determines the number of clusters automatically while quantifying the uncertainty of each samples assignment. Spike-and-slab priors identify discriminative features, and a Bayesian neural network estimates probabilistic co-occurrence between microbial species and host omics features. To isolate the effect of latent geometry, we evaluate two variants that are identical except for their latent space: DBayesCM-vMF constrains the latent to the unit hypersphere and applies a von Mises-Fisher mixture, while DBayesCM-GMM uses a Euclidean latent space and a Gaussian mixture. On simulated data, the hyperspherical variant recovered the correct number of clusters, whereas the Euclidean variant over-segmented, demonstrating that the latent geometry affects cluster recovery. Applied to colon, breast, and kidney cancer cohorts spanning metagenomics, host metabolomics, RNA-seq, and miRNA data, and to an obstructive sleep apnea model, DBayesCM ranked consistently among the existing methods while uniquely combining data-driven cluster-number determination, sample-level uncertainty, and interpretable feature selection within a single framework. DBayesCM reveals conditional probabilistic co-occurrence between core microbial species and host omics features, enabling uncertainty-aware exploration of microbiome-host relationships across diverse diseases.

bioinformatics↗

bioRxiv2026

vDeepInsight: an injective three-dimensional voxel carrier for tabular-feature neighborhood learning

DeepInsight-style methods make tabular feature relationships accessible to convolutional networks by placing each feature at a fixed position on an image carrier. An open design question is how the carrier geometry should be constructed when feature neighborhoods themselves carry part of the signal. We introduce vDeepInsight, an injective three-dimensional (3D) voxel carrier that preserves feature neighborhoods more faithfully than matched two-dimensional (2D) carriers while keeping a one-to-one mapping from each feature to a single voxel. Genes are embedded with t-SNE or UMAP, assigned one-to-one to a sparse voxel grid by linear-sum assignment, and processed by a submanifold sparse 3D convolutional network. We evaluate the carrier on gene expression through four linked analyses. First, representation-quality metrics show that 3D layouts reduce gene-neighborhood distortion relative to matched 2D layouts before any model is trained. Second, controlled synthetic tasks show that a sparse 3D convolution can exploit this preserved locality, but only when the supervised signal is constructed to depend on co-located genes and the receptive field spans adjacent voxels. Third, on real omics tasks the 3D carrier matches or exceeds tuned tabular baselines and consistently exceeds matched 2D carriers; the margin is small on marker-type classification, where individual genes already carry much of the label (tissue, lineage and cancer-type classification), and larger on program-type tasks, where the target depends on coordinated, pathway-level multi-gene activity (drug-response regression, TCGA immunogenomic-context regression and mechanism-of-action classification). Fourth, because the assignment is injective, voxel attribution maps directly back to genes, enabling gene-level attribution and pathway-level functional interpretation without voxel-to-gene deconvolution. Overall, the added carrier dimension improves the fidelity of feature-neighborhood representation and translates this improvement into prediction gains that are largest when the signal is distributed across local gene programs rather than dominated by individual marker genes.

bioinformatics↗

bioRxiv2026

Training Strategy Optimization to Mitigate Shortcut Learning in Pan-Cancer Drug Response Prediction

BackgroundPrediction of in vivo drug response is a central challenge in precision medicine, but the scarcity of labeled clinical data still necessitates the use of large-scale cancer cell line resources for model training. Domain adaptation methods, which aim to transfer knowledge learned from a source domain (cell lines) to a target domain (patients) by aligning feature distributions across domains, are a promising approach to bridge the gap between in vitro models and in vivo patients. However, we observed that these methods can exhibit a significant discrepancy between pan-cancer evaluation metrics and cancer type-specific prediction accuracy. This performance gap warrants a detailed investigation into their underlying predictive characteristics. ResultsWe discovered that cancer-type-specific class imbalances in training data can lead domain adaptation models to engage in shortcut learning, where they primarily discriminate between cancer types rather than capturing the actual biological determinants of drug sensitivity. To address this, we propose a strategy of combining two approaches: (1) excluding cancer types causing imbalance from the training data, and (2) adjusting class balance through oversampling and class weighting while retaining cancer types causing the imbalance. Among all configurations tested in conjunction with the CODE-AE (Context-aware Deconfounding AutoEncoder) framework, the combination of moderate oversampling (30% non-responder ratio) with class weighting achieved the best performance, significantly improving prediction accuracy in 5 out of 11 external patient cohorts from TCGA and GEO. ConclusionsOur findings demonstrate that appropriate class imbalance correction--rather than wholesale exclusion of imbalanced cancer subtypes--enables effective utilization of biologically relevant information shared across cancer types for drug response prediction. This study highlights the critical importance of jointly optimizing training data composition and class balance adjustment strategies in developing robust pan-cancer drug response prediction models for precision medicine applications. HighlightsO_LIIdentified a critical discrepancy in current domain adaptation models for drug response prediction: high pan-cancer accuracy often masks poor performance within specific cancer types. C_LIO_LIRevealed the root cause as "shortcut learning," where models tend to distinguish between cancer tissue types (hematological vs. solid) rather than learning individual drug sensitivity. C_LIO_LIDiscovered severe class imbalance in training data, with hematological cell lines being disproportionately drug-responsive across multiple chemotherapeutics. C_LIO_LIProposed an architecture-agnostic fix using the CODE-AE framework: moderate oversampling (30% minority ratio) combined with class weighting. C_LIO_LIDemonstrated significant improvements in 5 of 11 external patient cohorts, showing that correcting class bias is more effective than simply excluding problematic data. C_LI

bioinformatics↗

bioRxiv2026

Dynamic multimodal survival prediction in multiple myeloma integrating gene expression, longitudinal laboratories, and treatment history

Prognostic stratification in multiple myeloma (MM) relies on staging systems that assign patients to fixed categories at diagnosis and discard the temporal information that accumulates during treatment. We developed a dynamic multimodal framework that predicts residual overall survival using observation windows ranging from 1 to 18 months post-diagnosis. The model integrates DeepInsight-transformed gene expression representation, longitudinal laboratory measurement trajectories across 10 analytes, and treatment history for three drug classes through an adaptive fusion mechanism that accounts for missing clinical observations. On the MMRF CoMMpass cohort (n = 752), five-fold cross-validation yielded a concordance index (C-index) of 0.773 {+/-} 0.024 and a time-dependent AUC at a 1-year prediction horizon (tdAUC1yr) of 0.789 {+/-} 0.021, outperforming all evaluated baseline methods including DeepSurv (0.633 {+/-} 0.095) and random survival forests (0.636 {+/-} 0.024) on matched cross-validation splits. Modality ablation identified longitudinal laboratory measurements as the strongest individual contributor (C-index 0.693); the DeepInsight spatial encoding of gene expression yielded higher discrimination than a multilayer perceptron (MLP) baseline operating on the same features (0.624 vs. 0.596). Kaplan-Meier analysis showed significant prognostic group separation at all primary landmarks (log-rank p < 0.001; hazard ratios 3.46-3.93). A distilled student model retaining only the DeepInsight representation and five baseline clinical features achieved C-index 0.672 and tdAUC1yr 0.740 on an independent microarray cohort (GSE24080, n = 507) without retraining. Interpretability analysis identified prognostic associations consistent with established myeloma biology, including ubiquitin-proteasome pathway genes, endoplasmic reticulum stress markers, and Interferon Alpha Response pathway enrichment.

bioinformatics↗

bioRxiv2025

DNA Methylation Ageing Atlas Across 17 Human Tissues

Aging involves widespread epigenetic remodeling across tissues, yet the nature and consistency of these changes remain unclear. We conducted a meta-analysis of more than 15,000 human methylomes spanning 17 tissues, identifying both conserved and tissue-specific aging signatures. We examined linear changes via differentially methylated positions, variability shifts via variably methylated positions, and Shannon-entropy to capture methylation disorder. Network analysis revealed fragile co-methylation modules largely resistant to beneficial perturbation. Key disruptors, including PCDHGA1, MEST, HDAC4, and HOX genes, exacerbated aging signals across tissues. Notably, a resilient module enriched for NAD{square} salvage metabolism supports therapeutic targeting of NAD{square} in aging. PCDHGA1 emerged as a conserved cross-tissue driver, suggesting protocadherin-mediated adhesion plays a broader role in maintaining structural and signaling stability in multiple organ systems. Our open-access atlas provides a foundational resource for dissecting the molecular architecture of human aging and identifying testable targets for intervention, biomarkers, and translational epigenetic therapies.

bioinformatics↗

bioRxiv2025

scHDeepInsight: A Hierarchical Deep Learning Framework for Precise Immune Cell Annotation in Single-Cell RNA-seq Data

Immune cell classification from single-cell RNA sequencing (scRNA-seq) presents significant challenges due to complex hierarchical relationships among cell types. We introduce scHDeepInsight, a deep learning framework that extends our previous scDeepInsight model by integrating a biologically-informed classification architecture with an adaptive hierarchical focal loss. The framework leverages our established method of transforming gene expression data into two-dimensional structured images for use with convolutional neural networks by effectively capturing both global and fine-grained transcriptomic features, overcoming the limitations of flat classification approaches that ignore hierarchical relationships between cell types. scHDeepInsight dynamically adjusts loss contributions to balance performance across the hierarchy levels. It also employs STACAS batch correction, robust random masking, and SHAP-based interpretability to enhance prediction accuracy and biological insight. Comprehensive benchmarking across seven diverse tissue datasets shows scHDeepInsight achieves an average accuracy of 93.2%, representing a 5.1 percentage point improvement over current state-of-the-art methods. The model successfully distinguishes 50 distinct immune cell subtypes with high accuracy, demonstrating proficiency for identifying rare and closely related cell subtypes. These advantages make scHDeepInsight a robust tool for high-resolution immune cell subtype characterization, well suited for detailed immune profiling in immunological studies.

bioinformatics↗

bioRxiv2024

VBayesMM: Variational Bayesian neural network to prioritize important relationships of high-dimensional microbiome multiomics data

The analysis of high-dimensional microbiome multiomics datasets is crucial for understanding the complex interactions between microbial communities and host physiological states across health and disease conditions. Despite their importance, current methods such as the microbe-metabolite vectors (MMvec) approach, often fail to efficiently identify keystone species. This arises from the vast dimensionality of metagenomics data which complicates the inference of significant relationships, particularly the estimation of co-occurrence probabilities between microbes and metabolites. Here we propose the variational Bayesian microbiome multiomics (VBayesMM) approach, which enhances MMvec by incorporating a spike-and-slab prior within a Bayesian neural network. This allows VBayesMM to rapidly and precisely identify crucial microbial species, improving the accuracy of estimated co-occurrence probabilities between microbes and metabolites, while also robustly managing the uncertainty inherent in high-dimensional data. Moreover, we have implemented variational inference to address computational bottlenecks, enabling scalable analysis across extensive multiomics datasets. Our comparative evaluation of large-scale human and mouse multiomics datasets demonstrates that VBayesMM not only outperforms existing methods in accuracy but also provides a scalable solution for analyzing massive datasets. VBayesMM enhances the interpretability of the Bayesian neural network by identifying a core set of influential microbial species, thus facilitating a deeper understanding of their probabilistic relationships with the host.

bioinformatics↗

bioRxiv2023

Multi-representation DeepInsight: an improvement on tabular data analysis

Tabular data analysis is a critical task in various domains, enabling us to uncover valuable insights from structured datasets. While traditional machine learning methods have been employed for feature engineering and dimensionality reduction, they often struggle to capture the intricate relationships and dependencies within real-world datasets. In this paper, we present Multi-representation DeepInsight (abbreviated as MRep-DeepInsight), an innovative extension of the DeepInsight method, specifically designed to enhance the analysis of tabular data. By generating multiple representations of samples using diverse feature extraction techniques, our approach aims to capture a broader range of features and reveal deeper insights. We demonstrate the effectiveness of MRep-DeepInsight on single-cell datasets, Alzheimers data, and artificial data, showcasing an improved accuracy over the original DeepInsight approach and machine learning methods like random forest and L2-regularized logistic regression. Our results highlight the value of incorporating multiple representations for robust and accurate tabular data analysis. By embracing the power of diverse representations, MRep-DeepInsight offers a promising avenue for advancing decision-making and scientific discovery across a wide range of fields.

bioinformatics↗

bioRxiv2023

scDeepInsight: a supervised cell-type identification method for scRNA-seq data with deep learning

Annotation of cell-types is a critical step in the analysis of single-cell RNA sequencing (scRNA-seq) data that allows the study of heterogeneity across multiple cell populations. Currently this is most commonly done using unsupervised clustering algorithms, which project single-cell expression data into a lower dimensional space and then cluster cells based on their distances from each other. However, as these methods do not use reference datasets, they can only achieve a rough classification of cell-types, and it is difficult to improve the recognition accuracy further. To effectively solve this issue we propose a novel supervised annotation method, scDeepInsight. The scDeepInsight method is capable of performing manifold assignments. It is competent in executing data integration through batch normalization, performing supervised training on the reference dataset, doing outlier detection and annotating cell-types on query datasets. Moreover, it can help identify active genes or marker genes related to cell-types. The training of the scDeepInsight model is performed in a unique way. Tabular scRNA-seq data are first converted to corresponding images through the DeepInsight methodology. DeepInsight can create a trainable image transformer to convert non-image RNA data to images by comprehensively comparing interrelationships among multiple genes. Subsequently, the converted images are fed into convolutional neural networks (CNNs) such as EfficientNet-b3. This enables automatic feature extraction to identify the cell-types of scRNA-seq samples. We benchmarked scDeepInsight with six other mainstream cell annotation methods. The average accuracy rate of scDeepInsight reached 87.5%, which is more than 7% higher compared with the state-of-the-art methods.

bioinformatics↗

bioRxiv2022

DeepInsight-3D for precision oncology: an improved anti-cancer drug response prediction from high-dimensional multi-omics data with convolutional neural networks

Modern oncology offers a wide range of treatments and therefore choosing the best option for particular patient is very important for optimal outcomes. Multi-omics profiling in combination with AI-based predictive models have great potential for streamlining these treatment decisions. However, these encouraging developments continue to be hampered by very high dimensionality of the datasets in combination with insufficiently large numbers of annotated samples. In this study, we propose a novel deep learning-based method to predict patient-specific anticancer drug response from three types of multiomics data. The proposed DeepInsight-3D approach relies on structured data-to-image conversion that then allows use of convolutional neural networks, which are particularly robust to high dimensionality of the inputs while retaining capabilities to model highly complex relationships between variables. Of particular note, we demonstrate that in this formalism additional channels of an image can be effectively used to accommodate data from different omics layers while explicitly encoding the connection between them. DeepInsight-3D was able to outperform two other state-of-the-art methods proposed for this task. These advances can facilitate the development of better personalized treatment strategies for different cancers in the future.

bioinformatics↗

bioRxiv2020

DeepInsight-FS: Selecting features for non-image data using convolutional neural network

Identifying smaller element or gene subsets from biological or other data types is an essential step in discovering underlying mechanisms. Statistical machine learning methods have played a key role in revealing gene subsets. However, growing data complexity is pushing the limits of these techniques. A review of the recent literature shows that arranging elements by similarity in image-form for a convolutional neural network (CNN) improves classification performance over treating them individually. Expanding on this, here we show a pipeline, DeepInsight-FS, to uncover gene subsets of clinical relevance. DeepInsight-FS converts non-image samples into image-form and performs element selection via CNN. To our knowledge, this is the first approach to employ CNN for element or gene selection on non-image data. A real world application of DeepInsight-FS to publicly available cancer data identified gene sets with significant overlap to several cancer-associated pathways suggesting the potential of this method to discover biomedically meaningful connections.

bioinformatics↗

bioRxiv2018

An integrative machine learning approach for prediction of toxicity-related drug safety

Recent trends in drug development have been marked by diminishing returns of escalating costs and falling rate of new drug approval. Unacceptable drug toxicity is a substantial cause of drug failure during clinical trials as well as the leading cause of drug withdraws after release to market. Computational methods capable of predicting these failures can reduce waste of resources and time devoted to the investigation of compounds that ultimately fail. We propose an original machine learning method that leverages identity of drug targets and off-targets, functional impact score computed from Gene Ontology annotations, and biological network data to predict drug toxicity. We demonstrate that our method (TargeTox) can distinguish potentially idiosyncratically toxic drugs from safe drugs and is also suitable for speculative evaluation of different target sets to support the design of optimal low-toxicity combinations.\n\nSummary blurbPrediction of toxicity-related drug clinical trial failures, withdrawals from market and idiosyncratic toxicity risk by combining biological network analysis with machine learning.

systems biology↗

bioRxiv2017

Burly1 is a mouse QTL for lean body mass that maps to a 0.8-Mb region on chromosome 2

Our goal was to fine map a mouse QTL for lean body mass (Burly1) using information from several populations including newly created congenic mice derived from the B6 (host) and 129 (donor) strains. The results from each mapping population were concordant and showed that Burly1 is likely a single QTL in a 0.8-Mb region at 151.9-152.7 Mb (rs33197365 to rs3700604) on mouse chromosome 2. Results from mice of all the mapping populations we studied including intercrossed, backcrossed, consomic, and congenic strains indicate that lean body mass was increased by the B6-derived allele relative to the 129-derived allele. We determined that the congenic region harboring Burly1 contains 26 protein-coding genes, 11 noncoding RNA elements (e.g., lncRNA), and 4 pseudogenes, with 1949 predicted functional variants. The effect of the Burly1 locus on lean body weight was apparent at all ages measured and did not affect food intake or locomotor activity. However, congenic mice with the B6-allele produced more heat per kilogram of lean body weight than did controls, pointing to a genotype effect on lean mass metabolism. These results show the value of integrating information from several mapping populations to refine the map location of body composition QTLs.

genetics↗

bioRxiv2017

Adiposity QTL Adip20 decomposes into at least four loci when dissected using congenic strains

An average mouse in midlife weighs between 25 and 30 g, with about a gram of tissue in the largest adipose depot (gonadal), and the weight of this depot differs between inbred strains. Specifically, C57BL/6ByJ mice have heavier gonadal depots on average than do 129P3/J mice. To understand the genetic contributions to this trait, we mapped several quantitative trait loci (QTLs) for gonadal depot weight in an F2 intercross population. Our goal here was to fine-map one of these QTLs, Adip20 (formerly Adip5), on mouse chromosome 9. To that end, we analyzed the weight of the gonadal adipose depot from newly created congenic strains. Results from the sequential comparison method indicated at least four rather than one QTL; two of the QTLs were less than 0.5 Mb apart, with opposing directions of allelic effect. Different types of evidence (missense and regulatory genetic variation, human adiposity/body mass index orthologues, and differential gene expression) implicated numerous candidate genes from the four QTL regions. These results highlight the value of mouse congenic strains and the value of this sequential method to dissect challenging genetic architecture.

genetics↗