Biology subjects

Roshchupkin, G. V.

Publications and source records attributed to Roshchupkin, G. V..

7 recordsLinked to original sources

bioRxiv2025

Genetic Insights of Image-Based Traits: Analysis Pipeline for AI-based Phenotyping, Combined-GWAS, and Federated Learning with Application to the Human Face

Biological images capture rich morphological information and understanding their genetic basis is crucial for elucidating underlying molecular mechanisms, relevant in biomedical, evolutionary, and forensic research and applications. However, progress in understanding the genetic basis of image-based complex traits is hindered by key limitations in the current methodology, regarding both, the degree of which the large image complexity is captured with the phenotyping methods and how the analysis methods deal with the underlying large genetic complexity. Here, we present a robust, scalable, privacy-preserving computer pipeline for analysing image-based complex traits and unveiling their genetic basis by integrating (i) AI-based phenotyping for large-scale extraction of image-based traits across cohorts in a federated learning framework, without sharing individual images; (ii) Combined-GWAS (C-GWAS) for identifying genetic variants underlying the numerous AI-derived image-based traits; and (iii) Explainable AI (XAI)-based image visualization of the identified genetic association effects. In its first application to the example of the human face, we analysed 3D facial images and genomic data from two European cohorts (N=7,309), extracted 200 AI-derived facial traits, identified 43 significantly face associated genetic loci, including 12 novel ones, and replicated 70% of them in an independent European dataset (N=8,246). XAI-based visualization of the identified genetic effects shows the involvement of many of these loci in different parts of the face. Our study provides a privacy-aware and extensible pipeline for investigating the genetic basis of image-based complex traits implemented in a computationally efficient python package available at GitLab and its first application yielded new insights into the genetic architecture of facial shape variation.

genetics↗

bioRxiv2024

Detecting Genetic Interactions with Visible Neural Networks

Non-linear interactions among single nucleotide polymorphisms (SNPs), genes, and pathways play an important role in human diseases, but identifying these interactions is a challenging task. Neural networks are state-of-the-art predictors in many domains due to their ability to analyze big data and model complex patterns, including non-linear interactions. In genetics, visible neural networks are gaining popularity as they provide insight into the most important SNPs, genes and pathways for prediction. Visible neural networks use prior knowledge (e.g. gene and pathway annotations) to define the connections between nodes in the network, making them sparse and interpretable. Currently, most of these networks provide measures for the importance of SNPs, genes, and pathways but lack details on the nature of the interactions. In this paper, we explore different methods to detect non-linear interactions with visible neural networks. We adapted and sped up existing methods, created a comprehensive benchmark with simulated data from GAMETES and EpiGEN, and demonstrated that these methods can extract multiple types of interactions from trained visible neural networks. Finally, we applied these methods to a genome-wide case-control study of inflammatory bowel disease and found high consistency of the epistasis pairs candidates between the interpretation methods. The follow-up association test on these candidate pairs identified seven significant epistasis pairs.

bioinformatics↗

bioRxiv2024

Novel multi-omics deconfounding variational autoencoders can obtain meaningful disease subtyping

Unsupervised learning, particularly clustering, plays a pivotal role in disease subtyp- ing and patient stratification, especially with the abundance of large-scale multi-omics data. Deep learning models, such as variational autoencoders (VAEs), can enhance clustering algorithms by leveraging inter-individual heterogeneity. However, the impact of confounders - external factors unrelated to the condition, e.g. batch effect or age - on clustering is often overlooked, introducing bias and spurious biological conclusions. In this work, we introduce four novel VAE-based deconfounding frameworks tailored for clustering multi-omics data. These frameworks effectively mitigate confounding effects while preserving genuine biological patterns. The deconfounding strategies employed include: i) removal of latent features correlated with confounders ii) a conditional variational autoencoder, iii) adversarial training, and iv) adding a regularization term to the loss function. Using real-life multi-omics data from TCGA, we simulated various confounding effects (linear, non-linear, categorical, mixed) and assessed model performance across 50 repetitions based on reconstruction error, clustering stability, and deconfounding efficacy. Our results demonstrate that our novel models, particularly the conditional multi-omics VAE (cXVAE), successfully handle simulated confounding effects and recover biologically-driven clustering structures. cXVAE accurately identifies patient labels and unveils meaningful pathological associations among cancer types, validating deconfounded representations. Furthermore, our study suggests that some of the proposed strategies, such as adversarial training, prove insufficient in confounder removal. In summary, our study contributes by proposing innovative frameworks for simultaneous multi-omics data integration, dimensionality reduction, and deconfounding in clustering. Benchmarking on open-access data offers guidance to end-users, facilitating meaningful patient stratification for optimized precision medicine.

genomics↗

bioRxiv2023

mEthAE: an Explainable AutoEncoder for methylation data

1In the quest to unravel the mysteries of our epigenetic landscape, researchers are continually challenged by the relationships among CpG sites. Traditional approaches are often limited by the immense complexity and high dimensionality of DNA methylation data. To address this problem, deep learning algorithms, such as autoencoders, are increasingly applied to capture the complex patterns and reduce dimensionality into latent space. In this pioneering study, we introduce an innovative chromosome-wise autoencoder, termed mEthAE, specifically designed for the interpretive reduction of methylation data. mEthAE achieves an impressive 400-fold reduction in data dimensions without compromising on reconstruction accuracy or predictive power in the latent space. In attempt to go beyond mere data compression, we developed a perturbation-based method for interpretation of latent dimensions. Through our approach we identified clusters of CpG sites that exhibit strong connections across all latent dimensions, which we refer to as global CpGs. Remarkably, these global CpGs are more frequently highlighted in epigenome-wide association studies (EWAS), suggesting our methods ability to pinpoint biologically significant CpG sites. Our findings reveal a surprising lack of correlation patterns, or even physical proximity on the chromosome among these connected CpGs. This leads us to propose an intriguing hypothesis: our autoencoder may be detecting complex, long-range, non-linear interaction patterns among CpGs. These patterns, largely uncharacterised in current epigenetic research, hold the potential to shed new light on our understanding of epigenetics. In conclusion, this study not only showcases the power of autoencoders in untangling the complexities of epigenetic data but also opens up new avenues for understanding the hidden connections within CpGs. O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=168 SRC="FIGDIR/small/549496v2_ufig1.gif" ALT="Figure 1"> View larger version (31K): org.highwire.dtl.DTLVardef@16c5076org.highwire.dtl.DTLVardef@16b455org.highwire.dtl.DTLVardef@9948e8org.highwire.dtl.DTLVardef@181914b_HPS_FORMAT_FIGEXP M_FIG C_FIG

bioinformatics↗

bioRxiv2023

Phenotype prediction using biologically interpretable neural networks on multi-cohort multi-omics data

Integrating multi-omics data into predictive models has the potential to enhance accuracy, which is essential for precision medicine. In this study, we developed interpretable predictive models for multi-omics data by employing neural networks informed by prior biological knowledge, referred to as visible networks. These neural networks offer insights into the decision-making process and can unveil novel perspectives on the underlying biological mechanisms associated with traits and complex diseases. We tested the performance, interpretability, and generalizability for inferring smoking status, subject age and LDL levels using genome-wide RNA-expression and CpG methylation data from blood of the BIOS consortium(4 population cohorts, N_total=2940). In a cohort-wise cross validation setting, the consistency of the diagnostic performance and interpretation was assessed. Performance was consistently high for predicting smoking status with an overall mean AUC of 0.95 (95% CI, 0.90 - 1.00) and interpretation revealed the involvement of well-replicated genes such as AHRR, GPR15 and LRRN3. LDL-level predictions only generalized in a single cohort with an R2 of 0.07 (95% CI, 0.05 - 0.08). Age was infered with a mean error of 5.16 (95% CI, 3.97 - 6.35) years with the genes COL11A2, AFAP1, OTUD7A, PTPRN2, ADARB2 and CD34 consistently predictive. In general, we found that using multi-omics networks improved performance, stability and generalizability compared to interpretable single omic networks. We believe that visible neural networks have great potential for multi-omics analysis; they combine multi-omic data elegantly, are interpretable, and generalize well to data from different cohorts.

genomics↗

bioRxiv2020

GenNet framework: interpretable neural networks for phenotype prediction

Deep learning is rarely used in population genomics because of the computational burden and challenges in interpreting neural networks. Here, we propose GenNet, a novel open-source deep learning framework for predicting phenotypes from genetic variants. In this framework, interpretable and memory-efficient neural network architectures are constructed by embedding biological knowledge from public databases, resulting in neural networks that contain only biological plausible connections. We applied the framework to seventeen phenotypes from a case-control study, a population-based study and the UK Biobank. Interpreting the networks revealed well-replicated genes such as HERC2 and OCA2 for hair and eye color and novel genes such as ZNF773 and PCNT for schizophrenia. Additionally, the framework obtained an AUC of 0.74 in the held-out test set and identified ubiquitin mediated proteolysis, endocrine system and viral infectious diseases as most predictive biological pathways for schizophrenia. GenNet is a freely available, end-to-end deep learning framework that allows researchers to develop and use interpretable neural networks to obtain novel insights into the genetic architecture of complex traits and diseases.

bioinformatics↗

bioRxiv2018

Genetic Determinants of Cortical Structure (Thickness, Surface Area and Volumes) among Disease Free Adults in the CHARGE Consortium

Cortical thickness, surface area and volumes (MRI cortical measures) vary with age and cognitive function, and in neurological and psychiatric diseases. We examined heritability, genetic correlations and genome-wide associations of cortical measures across the whole cortex, and in 34 anatomically predefined regions. Our discovery sample comprised 22,822 individuals from 20 cohorts within the Cohorts for Heart and Aging Research in Genomic Epidemiology (CHARGE) consortium and the United Kingdom Biobank. Significant associations were replicated in the Enhancing Neuroimaging Genetics through Meta-analysis (ENIGMA) consortium, and their biological implications explored using bioinformatic annotation and pathway analyses. We identified genetic heterogeneity between cortical measures and brain regions, and 161 genome-wide significant associations pointing to wnt/{beta}-catenin, TGF-{beta} and sonic hedgehog pathways. There was enrichment for genes involved in anthropometric traits, hindbrain development, vascular and neurodegenerative disease and psychiatric conditions. These data are a rich resource for studies of the biological mechanisms behind cortical development and aging.

genetics↗