Biology subjects

Sokolova, K.

Publications and source records attributed to Sokolova, K..

6 recordsLinked to original sources

bioRxiv2026

Multi-modal tissue-aware graph neural network for in silico genetic discovery

Understanding how perturbations influence gene function in a tissue-specific manner is key to uncovering novel drug targets. However, current computational approaches emphasize global network or sequence-derived features over context-driven dependencies. We introduce Mahi, a scalable and interpretable graph neural network framework that learns gene representations by integrating chromatin accessibility, transcription factor binding, histone modifications, and protein structure features in tissue-specific contexts. Through pretraining on tissue-specific network topologies followed by multi-modal feature integration, Mahi learns context-aware gene embeddings across 290 tissues and cell-types. Mahi outperforms sequence-based models in predicting gene essentiality across 1,183 cancer cell lines, demonstrating the advantage of integrating molecular context and functional connectivity. The learned embedding space reveals tissue-specific functional organization, with genes forming distinct clusters reflecting their context-dependent roles. In silico gene knockout perturbations demonstrate Mahis ability to model intricate perturbation responses, identifying disease-relevant pathways and therapeutic targets. Together, these results demonstrate Mahi as a foundation for modeling tissue-specific gene function and perturbation responses, enabling applications in precision medicine, therapeutic target discovery, and prediction of context-dependent genetic vulnerabilities. All embeddings and the framework are publicly available to facilitate use by the scientific community.

genomics↗

bioRxiv2026

Gene-centered representation of coding and regulatory variation enables outcome prediction

Integrating coding and regulatory variation into unified, interpretable representations remains a challenge in functional genomics. Current approaches either focus on common variants or analyze individual variants in isolation, missing the cumulative, cell-type-specific impact of both coding and noncoding variants on each gene. We present Volaria, a computational framework that integrates coding and regulatory genetic variation into unified, gene-centered representations for disease outcome prediction from whole-genome sequencing. Volaria leverages deep learning models to capture variant effects on cell-type-specific gene expression and integrates them with AI-predicted exonic variant pathogenicity to produce representations that capture the cumulative effect of genome-wide rare and common variation. Applied to whole genomes of individuals with rare glomerular diseases, Volaria predicts individual outcomes directly from germline sequence, demonstrating that structured, cell-type-aware representations capture predictive signals beyond population-based polygenic risk scores and unstructured representations. Importantly, the framework identifies context-specific biological mechanisms, providing interpretability that can be aligned with clinical measurements. By encoding genome-wide variation into compact and biologically grounded representations, Volaria provides a scalable foundation for genome interpretation and individualized outcome modeling from germline sequence, complementing phenotypic and clinical information in the future integrative frameworks.

genomics↗

bioRxiv2025

An Evidence-Grounded Research Assistant for Functional Genomics and Drug Target Assessment

The growing availability of biological data resources has transformed research, yet their effective use remains challenging: selecting appropriate sources requires domain knowledge, data are fragmented across databases, and synthesizing results into reliable conclusions is labor-intensive. Although large language models promise to address these barriers, their impact in biomedicine has been limited by unsupported statements, incorrect claims, and lack of provenance. We introduce Alvessa, an evidence-grounded agentic research assistant designed around verifiability. Alvessa integrates entity recognition, orchestration of pre-validated biological tools, and data-constrained answer generation with statement-level verification against retrieved records, explicitly flagging unsupported claims and guiding revision when reliability criteria are not met. We evaluate Alvessa on dbQA from LAB-Bench and GenomeArena, a benchmark of 720 questions spanning gene and variant annotation, pathways, molecular interactions, miRNA targets, drug-target evidence, protein structure, and gene-phenotype associations. Alvessa substantially improves accuracy relative to general-purpose language models and performs comparably to coding-centric agents while producing fully traceable outputs. Using adversarial perturbations, we show that detection of fabricated statements depends critically on access to retrieved evidence. We further demonstrate application to drug discovery, where evidence-grounded synthesis enables identification of candidate targets missed or misattributed by literature-centered reasoning alone. Alvessa and GenomeArena are released to the community to support reproducible, verifiable AI-assisted biological research.

bioinformatics↗

bioRxiv2025

CLONAL SELECTION SUPPORTED BY SINGLE CELL DNA SEQUENCING REVEALS HORMONAL ADAPTATION AND RESISTANCE IN LOCALLY ADVANCED BREAST CANCER DURING NEOADJUVANT AROMATASE INHIBITION

BackgroundThe aromatase inhibitors (AI) letrozole and exemestane, are often used in sequence in targeting ER+ breast cancers. However resistance to AI poses a major barrier to sustained clinical benefit, while the biological mechanisms underlying the phenomenon remain largely unknown. In this study, we build on our clinical NeoLetExe trial, with the aim to investigate the molecular basis of resistance to AI, by analysing subclonal evolutionary dynamics during sequential treatment. MethodsWe use whole-exome sequencing (WES) data from 11 ER+ breast cancer patients and 3 timepoints of the Neoletexe trial to reconstruct cancer cell fraction-based subclonal composition. Single-cell DNA sequencing from matched tumour samples is used for validating identified clones and variants. Subclonal variants were annotated to genes by integrating evidence from public data and ExpectoSc. Pathway enrichment analysis using Human Base was conducted. ResultsHigher cancer cell fraction clone trajectories were significantly associated with reduced treatment response (p = 0.023). Clones reconstructed by WES were validated at 81% using single-cell DNA sequencing. Clones resistant to both letrozole and exemestane demonstrated PIK3CA/AKT/mTOR signaling activation, KRAS pathway dysregulation, hedgehog signaling, and androgen receptor pathways, alongside extensive immune activation and metabolic reprogramming. Drug-specific resistance patterns showed exemestane-resistant clones enriched for epigenetic control and miRNA-mediated silencing, while letrozole-resistant clones displayed metabolic dysregulation but notably lacked immune pathway activation. In contrast, treatment-sensitive clones maintained coordinated cell cycle control, preserved DNA damage responses, and retained immune signaling capacity. Analysis of FDA-approved breast cancer targets identified actionable alterations in PIK3CA (4 patients) and AKT1 (1 patient) that persisted through AI treatment, with RNA expression analysis revealing 48 additional therapeutic targets spanning PI3K/AKT/mTOR, CDK4/6, DNA repair (BRCA1/2, ATM), and immune checkpoint pathways. ConclusionWES-based cancer cell fraction analysis successfully captured subclonal evolutionary trajectories during AI treatment, revealing drug-specific mechanisms and identifying key molecular players in endocrine therapy resistance. This work establishes a framework for precision oncology approaches by providing actionable therapeutic targets and advancing our understanding of resistance mechanisms to improve clinical outcomes in sequential AI therapy.

cancer biology↗

bioRxiv2025

Variant-resolved prediction of context-specific isoform variation with a graph-based attention model

In eukaryotes, most genes produce multiple transcript isoforms that diversify the transcriptome and proteome, serving as a key mechanism of functional regulation. Genetic variation can disrupt the RNA processing signals that shape isoform structure and abundance, yet modeling these effects at full-length isoform resolution remains challenging due to the complexity of transcript regulation. Here, we introduce Otari, an attention-based graph neural network framework trained on the human genomic sequence and long-read transcriptomes across 30 tissue types and brain regions. Otari predicts tissue-specific differential isoform abundance by integrating sequence-derived epigenetic and post-transcriptional signals, enabling isoform-resolved variant effect interpretation. Applied to large-scale variant datasets including an autism cohort, Otari uncovers patterns of isoform dysregulation undetectable at the gene level, such as variant-driven perturbations in isoform abundance and microexon usage implicated in autism pathophysiology. We provide Otari as a resource for powering isoform-level analyses across tissues at scale.

bioinformatics↗

bioRxiv2024

Contrastive pre-training for sequence based genomics models

MotivationIn recent years deep learning has become one of the central approaches in a number of applications, including many tasks in genomics. However, as models grow in depth and complexity, they either require more data or a strategic initialization technique to improve performance. ResultsIn this project, we introduce cGen, a novel unsupervised, model-agnostic contrastive pretraining method for sequence-based models. cGen can be used before training to initialize weights, reducing the size of the dataset needed. It works through learning the intrinsic features of the reference genome and makes no assumptions on the underlying structure. We show that the embeddings produced by the unsupervised model are already informative for gene expression prediction and that the sequence features provide a meaningful clustering. We demonstrate that cGen improves model performance in various sequence-based deep learning applications, such as chromatin profiling prediction and gene expression. Our findings suggest that using cGen, particularly in areas constrained by data availability, could improve the performance of deep learning genomic models without the need to modify the model architecture. Contactogt@cs.princeton.edu Availability and implementationSource code available at github.com/ksenia007/cGen

genomics↗