Biology subjects

Akalin, A.

Publications and source records attributed to Akalin, A..

At least 19 recordsLinked to original sources

bioRxiv2026

An atlas-scale generative model for unified representation learning of bulk RNA-seq data

Public bulk RNA-seq repositories contain hundreds of thousands of samples, creating opportunities for large-scale representation learning, but integration across studies remains challenging because of heterogeneous annotations, experimental protocols, and technical variation. While pre-trained foundation models are now widely available for single-cell RNA-seq, comparable resources for bulk RNA-seq remain scarce, motivating a model that learns a unified, tissue-aware representation directly from bulk data. We trained a supervised variational autoencoder (VAE) on a compendium of 118,263 bulk RNA-seq samples that we assembled from TCGA, GTEx, and ARCHS4 and mapped to 42 tissue categories. The model classifies tissue of origin at 94.9% balanced accuracy (weighted F1 96.2%) and compresses 16,115 genes into a 121-dimensional latent space. Tissue identity is the primary organizing axis of the latent space, while source effects remain secondary. To assess the impact of data volume, we constructed training sets at three different scales (38K, 75K, and 118K samples). Our results demonstrated that reconstruction fidelity improved incrementally with each expansion of the dataset, but with diminishing returns. We validated the model on an independent cohort of 734 paediatric tumour samples from TARGET, achieving 84.6% agreement with the expected tissue of origin. The trained model and code are available at GitHub (https://github.com/BIMSBbioinfo/flexynesis_tissue_vae_manuscript) with an interactive web application.

bioinformatics↗

bioRxiv2024

Enhancing Biomarker-Based Oncology Trial Matching Using Large Language Models

Clinical trials are an essential component of drug development for new cancer treatments, yet the information required to determine a patients eligibility for enrollment is scattered in large amounts of unstructured text. Genomic biomarkers are especially important in precision medicine and targeted therapies, making them essential for matching patients to appropriate trials. Large language models (LLMs) offer a promising solution for extracting this information from clinical trial data, aiding both physicians and patients in identifying suitable matches. In this study, we explore various LLM strategies for extracting genetic biomarkers from oncology trials to improve patient enrollment rates. Our results show that open-source language models, when applied out-of-the-box, effectively capture complex logical expressions and structure genomic biomarkers in disjunctive normal form, outperforming closed-source models such as GPT-4 and GPT-3.5-Turbo. Furthermore, fine-tuning these open-source models with additional data significantly enhances their performance.

bioinformatics↗

bioRxiv2024

Flexynesis: A deep learning framework for bulk multi-omics data integration for precision oncology and beyond

Accurate decision making in precision oncology depends on integration of multimodal molecular information, such as the genetic data, gene expression, protein abundance, and epigenetic measurements. Deep learning methods facilitate integration of heterogeneous datasets. However, almost all published deep learning-based bulk multi-omics integration methods have constrained usability. They suffer from lack of transparency, modularity, deployability, and are applicable exclusively to narrow tasks. To address these limitations, we introduce Flexynesis, a versatile tool designed with usability, and adaptability in mind. Flexynesis streamlines data processing, enforces structured data splitting, and ensures rigorous model evaluation. It offers unsupervised feature selection, different omics layer fusion options, and hyperparameter tuning. Users can choose from distinct architectures - fully connected networks, variational autoencoders, multi-triplet networks, graph neural networks, and cross-modality encoding networks. Each model is complemented with a straightforward input interface and standardized training, evaluation, and feature importance quantification methods, enabling easy incorporation into data integration pipelines. For improved user experience, Flexynesis supports features such as on-the-fly task determination and compatibility with regression, classification, and survival modeling. It accommodates multi-task prediction of a mixture of numerical/categorical outcome variables with a tolerance for missing labels. We also developed an extensive benchmarking pipeline, showcasing the tools capability across diverse real-life datasets. This toolset should make deep-learning based bulk multi-omics data integration in the context of clinical/pre-clinical data analysis and marker discovery more accessible to a wider audience with or without experience in deep-learning development. Flexynesis is available at https://github.com/BIMSBbioinfo/flexynesis and can be installed from https://pypi.org/project/flexynesis/. O_FIG O_LINKSMALLFIG WIDTH=199 HEIGHT=200 SRC="FIGDIR/small/603606v1_ufig1.gif" ALT="Figure 1"> View larger version (53K): org.highwire.dtl.DTLVardef@17eda85org.highwire.dtl.DTLVardef@13c89ecorg.highwire.dtl.DTLVardef@182f706org.highwire.dtl.DTLVardef@127d9d9_HPS_FORMAT_FIGEXP M_FIG O_FLOATNOGraphical Abstract:C_FLOATNO Summary of the Flexynesis data integration and analysis workflow. C_FIG

bioinformatics↗

bioRxiv2024

Oncogene inactivation-induced senescence facilitates tumor relapse

Oncogene-directed therapies can induce profound tumor regression in oncogene-addicted cancers such as BRAF-mutant melanoma and KRAS-driven pancreatic cancer, but their long-term benefit is often limited by resistance and early relapse. The mechanisms that allow residual cells to adapt, persist in a dormant state, and eventually fuel recurrence remain poorly understood. Here, we show that oncogene inactivation rapidly induces hallmark features of senescence together with a pro-inflammatory senescence-associated secretory phenotype (SASP). In vivo, oncogene inactivation-induced senescence (OIIS) predisposed tumors to relapse, characterized by polyploidy, chromosomal instability, and acquisition of alternative oncogenic pathways such as Mdm2 upregulation. Tumor microenvironment profiling by spectral flow cytometry revealed that relapse was associated with neovascularization and a shift from an immune-activated to an immunosuppressive milieu, indicating that senescent cells remodel their niche to promote regrowth. Importantly, OIIS features were also observed in human BRAFV600E melanoma cells treated with vemurafenib, confirming the clinical relevance of our findings. Together, our findings establish OIIS as a double-edged process: it initially restrains tumor growth but simultaneously creates conditions that favor recurrence. By defining the genetic, metabolic, cytogenetic, and microenvironmental hallmarks of OIIS, our study highlights adaptations to oncogene deprivation that limit the durability of targeted therapies.

cancer biology↗

bioRxiv2024

A single dose of cocaine rewires the 3D genome structure of midbrain dopamine neurons

Midbrain dopamine neurons (DNs) respond to a first exposure to addictive drugs and play key roles in chronic drug usage1-3. As the synaptic and transcriptional changes that follow an acute cocaine exposure are mostly resolved within a few days4,5, the molecular changes that encode the long-term cellular memory of the exposure within DNs remain unknown. To investigate whether a single cocaine exposure induces long-term changes in the 3D genome structure of DNs, we applied Genome Architecture Mapping and single nucleus transcriptomic analyses in the mouse midbrain. We found extensive rewiring of 3D genome architecture at 24 hours past exposure which remains or worsens by 14 days, outlasting transcriptional responses. The cocaine-induced chromatin rewiring occurs at all genomic scales and affects genes with major roles in cocaine-induced synaptic changes. A single cocaine exposure triggers extensive long-lasting changes in chromatin condensation in post-synaptic and post-transcriptional regulatory genes, for example the unfolding of Rbfox1 which becomes most prominent 14 days post exposure. Finally, structurally remodeled genes are most expressed in a specific DN sub-type characterized by low expression of the dopamine auto-receptor Drd2, a key feature of highly cocaine-sensitive cells. These results reveal an important role for long-lasting 3D genome remodelling in the cellular memory of a single cocaine exposure, providing new hypotheses for understanding the inception of drug addiction and 3D genome plasticity.

genomics↗

bioRxiv2024

DNA virus infections shape transposable elements activity in vitro and in vivo

Transposable elements (TEs) are implicated in a variety of processes including placental and preimplantation development and a variety of human diseases. TEs are known to be activated in the context of some viral infections, but the mechanisms and consequences are not understood. We show strong activation of TEs upon DNA virus infection, in particular the MLT- and THE1-class of LTR-containing retrotransposons as well as a subset of LINE-1-, Alu-elements and HERVs. Mechanistically, two key pathways induce TEs upregulation: inhibition of the KAP1/TRIM28 repressive complex by phosphorylation, and expression of the pioneer transcription factor double-homeobox 4 (DUX4), which is known to be involved in TE-induction during zygotic genome activation in embryonic development. DUX4 is induced by DNA viruses, it binds to TEs upon infection and analysis of genes adjacent to TEs shows pathways that are important for DNA virus infections. Analysis of knockdown, knockout and overexpression data reveal that almost all TEs expressed upon herpesviral infection are regulated by KAP1/TRIM28 and DUX4. Interestingly, analysis of single cell sequencing data from patients with DNA virus-associated cancers showed that in vivo TEs expression strongly correlates with virus infection, indicating a possible role in viral oncogenesis.

genomics↗

bioRxiv2023

VoltRon: A Spatial Omics Analysis Platform for Multi-Resolution and Multi-omics Integration using Image Registration

The growing number of spatial omic technologies have created a demand for computational tools capable of managing, storing, and analyzing spatial datasets with multiple modalities and spatial resolutions. Meanwhile, computer vision is becoming an integral part of processing spatial data readouts where image registration and spatial data alignment of tissue sections are essential prior to data integration. Hence, there is a need for computational platforms that analyze data across spatial datasets with diverse resolutions as well as those that manipulate and process images of microanatomical tissue structures. To this end, we have developed VoltRon, a novel R package for spatial omics analysis with a unique data structure that accommodates data readouts with many levels of spatial resolutions (i.e., multi-resolution) including regions of interest (ROIs), spots, single cells, and even subcellular entities such as molecules. To connect and integrate these spatially diverse omic profiles, VoltRon accounts for spatial organization of tissue blocks (samples), layers (sections) and assays given a multi-resolution collection of spatial data readouts. An easy-to-use computer vision toolbox, OpenCV, is fully embedded in VoltRon that allows users to both automatically and manually register spatial coordinates across adjacent layers for data transfer without the need for external software tools. VoltRon is implemented in the R programming language and is freely available at https://github.com/BIMSBbioinfo/VoltRon.

bioinformatics↗

bioRxiv2023

Leveraging large language models for data analysis automation

Data analysis is constrained by a shortage of skilled experts, particularly in biology, where detailed data interpretation is vital for understanding complex biological processes and developing new treatments and diagnostics. To address this, we developed mergen, an R package that leverages Large Language Models (LLMs) for data analysis code generation and execution. Our primary goal is to enable humans to conduct data analysis by simply describing their objectives and the desired analyses for specific datasets through clear text. Our approach improves code generation via specialized prompt engineering and error feedback mechanisms. In addition, our system can execute the data analysis workflows prescribed by the LLM providing the results of the data analysis workflow for human review. We evaluated the performance of this data analysis system using various data analysis tasks. Our evaluation revealed that while LLMs effectively generate code for some data analysis tasks, challenges remain in executable code generation, especially for complex data analysis tasks. Our study contributes to a better understanding of LLM capabilities and limitations, providing software infrastructure and practical insights for their effective integration into data analysis workflows.

bioinformatics↗

bioRxiv2023

Pathogenic mutations of human phosphorylation sites affect protein-protein interactions

Despite their lack of a defined 3D structure, intrinsically disordered regions (IDRs) of proteins play important biological roles. Many IDRs contain short linear motifs (SLiMs) that mediate protein-protein interactions (PPIs), which can be regulated by post-translational modifications like phosphorylation. 20% of pathogenic missense mutations are found in IDRs, and understanding how such mutations affect PPIs is essential for unraveling disease mechanisms. Here, we employed peptide-based interaction proteomics to investigate 36 disease-causing mutations affecting phosphorylation sites. Our results unveiled significant differences in interactomes between phosphorylated and non-phosphorylated peptides, often due to disrupted phosphorylation-dependent SLiMs. We focused on a mutation of a serine phosphorylation site in the transcription factor GATAD1, which causes dilated cardiomyopathy. We found that this phosphorylation site mediates interaction with 14-3-3 family proteins. Follow-up experiments revealed the structural basis of this interaction and suggest that 14-3-3 binding affects GATAD1 nucleocytoplasmic transport by masking a nuclear localisation signal. Our results demonstrate that pathogenic mutations of human phosphorylation sites can significantly impact protein-protein interactions, offering fresh insights into potential molecular mechanisms underlying pathogenesis.

systems biology↗

bioRxiv2023

Dysregulated Pulmonary Inflammatory Responses Exacerbate the Outcome of Secondary Aspergillosis Following Influenza

Inhalation of airborne conidia of the ubiquitous fungus Aspergillus fumigatus commonly occurs but invasive aspergillosis is rare except in profoundly immunocompromised persons. Severe influenza predisposes patients to invasive pulmonary aspergillosis by mechanisms that are poorly defined. Using a post-influenza aspergillosis model, we found that superinfected mice had 100% mortality when challenged with A. fumigatus conidia on days 2 and 5 (early stages) of influenza A virus infection but 100% survival when challenged on days 8 and 14 (late stages). Influenza-infected mice superinfected with A. fumigatus had increased levels of the pro-inflammatory cytokines and chemokines IL-6, TNF, IFN{beta}, IL-12p70, IL-1, IL-1{beta}, CXCL1, G-CSF, MIP-1, MIP-1{beta}, RANTES and MCP-1. Surprisingly, on histopathological analysis, superinfected mice did not have greater lung inflammation compared with mice infected with influenza alone. Mice infected with influenza had dampened neutrophil recruitment to the lungs following subsequent challenge with A. fumigatus, but only if the fungal challenge was executed during the early stages of influenza infection. However, influenza infection did not have a major effect on neutrophil phagocytosis and killing of A. fumigatus conidia. Moreover, minimal germination of conidia was seen on histopathology even in the superinfected mice. Taken together, our data suggest that the high mortality rate seen in mice during the early stages of influenza-associated pulmonary aspergillosis is multifactorial, with a greater contribution from dysregulated inflammation than microbial growth. ImportanceSevere influenza is a risk factor for fatal invasive pulmonary aspergillosis; however, the mechanistic basis for the lethality is unclear. Utilizing an influenza-associated pulmonary aspergillosis (IAPA) model, we found that mice infected with influenza A virus followed by A. fumigatus had 100% mortality when superinfected during the early stages of influenza but survived at later stages. While superinfected mice had dysregulated pulmonary inflammatory responses compared to controls, they had neither increased inflammation nor extensive fungal growth. Although influenza-infected mice had dampened neutrophil recruitment to the lungs following subsequent challenge with A. fumigatus, influenza did not affect the ability of neutrophils to clear the fungi. Our data suggest that the lethality seen in our model IAPA is multifactorial with dysregulated inflammation being a greater contributor than uncontrollable microbial growth. If confirmed in humans, our findings provide a rationale for clinical studies of adjuvant anti-inflammatory agents in the treatment of IAPA.

immunology↗

bioRxiv2023

Decoding Single-Cell Multiomics: scMaui - A Deep Learning Framework for Uncovering Cellular Heterogeneity in Presence of Batch Effects and Missing Data

The recent advances in high-throughput single-cell sequencing has significantly required computational models which can address the high complexity of single-cell multiomics data. Meticulous single-cell multiomics integration models are required to avoid biases towards a specific modality and overcome the sparsity. Batch effects obfuscating biological signals must also be taken into account. Here, we introduce a new single-cell multiomics integration model, Single-cell Multiomics Autoencoder Integration (scMaui) based on stacked variational encoders and adversarial learning. scMaui reduces the dimensionality of integrated data modalities to a latent space which outlines cellular heterogeneity. It can handle multiple batch effects independently accepting both discrete and continuous values, as well as provides varied reconstruction loss functions to cover various assays and preprocessing pipelines. We show that scMaui accomplishes superior performance in many tasks compared to other methods. Further downstream analyses also demonstrate its potential in identifying relations between assays and discovering hidden subpopulations.

bioinformatics↗

bioRxiv2022

A new type of transcriptional reprogramming by an IRF4 mutation in lymphoma

SUMMARY PARAGRAPHDisease-causing mutations in genes encoding transcription factors (TFs) are a recurrent finding in hematopoietic malignancies and might involve key regulators of lineage adherence and cellular differentiation1-3. Such mutations can affect TF-interactions with their cognate DNA-binding motifs4, 5. Whether and how TF-mutations impact upon the nature of binding to TF composite elements (CE) and influence their interaction with other TFs is unclear. Here, we report a new mechanism of TF alteration in human lymphomas with perturbed B cell identity. It is caused by a recurrent somatic missense mutation c.295T>C (p.Cys99Arg; p.C99R) targeting the center of the DNA-binding domain of Interferon Regulatory Factor 4 (IRF4), a key TF in immune cell-differentiation and -activation6, 7. IRF4-C99R fundamentally alters IRF4 DNA-binding, with loss-of-binding to canonical IRF motifs and neomorphic gain-of-binding to canonical and non-canonical IRF composite elements (CEs). Furthermore, IRF4-C99R thoroughly modifies IRF4 function, by blocking IRF4-dependent plasma cell induction, and up-regulating disease-specific genes in a non-canonical Activator Protein-1 (AP-1)-IRF-CE (AICE)-dependent manner. Our data explain how a single arginine mutation creates a complex switch of TF specificity and gene regulation. These data open the possibility of designing specific inhibitors to block the neomorphic, disease-causing DNA-binding activities of a mutant transcription factor.

cancer biology↗

bioRxiv2022

Multi-omics alleviates the limitations of panel-sequencing for cancer drug response prediction

Comprehensive genomic profiling using cancer gene panels has been shown to improve treatment options for a variety of cancer types. However, genomic aberrations detected via such gene panels dont necessarily serve as strong predictors of drug sensitivity. In this study, using pharmacogenomics datasets of cell lines, patient-derived xenografts, and ex-vivo treated fresh tumor specimens, we demonstrate that utilizing the transcriptome on top of gene panel features substantially improves drug response prediction performance in cancer.

bioinformatics↗

bioRxiv2022

The SPOC domain is a phosphoserine binding module that bridges transcription machinery with co- and post-transcriptional regulators

The heptarepeats of the C-terminal domain (CTD) of RNA polymerase II (Pol II) are extensively modified throughout the transcription cycle. The CTD coordinates RNA synthesis and processing by recruiting transcription regulation factors as well as RNA capping, splicing and 3end processing factors. The SPOC domain of PHF3 was recently identified as a new CTD reader domain specifically binding to phosphorylated serine-2 residues in adjacent CTD repeats. Here, we establish the SPOC domains of the human proteins DIDO, SHARP and RBM15 as phosphoserine binding modules that can act as CTD readers but also recognize other phosphorylated binding partners. We report the crystal structure of SHARP (SPEN) SPOC-CTD and identify the molecular determinants for its specific binding to phosphorylated serine-5. PHF3 and DIDO SPOC domains preferentially interact with the Pol II elongation complex, while RBM15 and SHARP SPOC domains engage with the m6A writer and reader proteins. Our findings establish the SPOC domain as a major interface between the transcription machinery and regulators of transcription and co-transcriptional processes.

biochemistry↗

bioRxiv2021

Identifying tumor cells at the single cell level

Tumors are highly complex tissues composed of cancerous cells, surrounded by a heterogeneous cellular microenvironment. Tumor response to treatments is governed by an interaction of cancer cell intrinsic factors with external influences of the tumor microenvironment. Disentangling the heterogeneity within a tumor is a crucial step in developing and utilization of effective cancer therapies. The single cell sequencing technology enables an effective molecular characterization of single cells within the tumor. This technology can help deconvolute heterogeneous tumor samples and thus revolutionize personalized medicine. However, a governing challenge in cancer single cell analysis is cell annotation, the assignment of a particular cell type or a cell state to each sequenced cell. One of the critical cell type annotation challenges is identification of tumor cells within single cell or spatial sequencing experiments.This is a critical limiting step for a multitude of research, clinical, and commercial applications. A reliable method addressing that challenge is a prerequisite for automatic annotation of histopathological data, profiled using multichannel immunofluorescence or spatial sequencing. Here, we propose Ikarus, a machine learning pipeline aimed at distinguishing tumor cells from normal cells at the single cell level. We have tested ikarus on multiple single cell datasets to ascertain that it achieves high sensitivity and specificity in multiple experimental contexts.

cancer biology↗

bioRxiv2021

Multi-omics and deep learning provide a multifaceted view of cancer

Cancer is a complex disease with a large financial and healthcare burden on society. One hallmark of the disease is the uncontrolled growth and proliferation of malignant cells. Unlike Mendelian diseases which may be explained by a few genomic loci, a deeper molecular and mechanistic understanding of the development of cancer is needed. Such an endeavor requires the integration of tens of thousands of molecular features across multiple layers of information encoded in the cells. In practical terms, this implies integration of multi omics information from the genome, transcriptome, epigenome, proteome, metabolome, and even micro-environmental factors such as the microbiome. Finding mechanistic insights and biomarkers in such a high dimensional space is a challenging task. Therefore, efficient machine learning techniques are needed to reduce the dimensionality of the data while simultaneously discovering complex but meaningful biomarkers. These markers then can lead to testable hypotheses in research and clinical applications. In this study, we applied advanced deep learning methods to uncover multi-omic fingerprints that are associated with a wide range of clinical and molecular features of tumor samples. Using these fingerprints, we can accurately classify different cancer types, and their subtypes. Non-linear multi-omic fingerprints can uncover clinical features associated with patient survival and response to treatment, ranging from chemotherapy to immunotherapy. In addition, multi-omic fingerprints may be deconvoluted into a meaningful subset of genes and genomic alterations to support clinically relevant decisions. Graphical Abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=123 SRC="FIGDIR/small/462364v1_ufig1.gif" ALT="Figure 1"> View larger version (44K): org.highwire.dtl.DTLVardef@4b53d0org.highwire.dtl.DTLVardef@16b9a96org.highwire.dtl.DTLVardef@1720212org.highwire.dtl.DTLVardef@1e5f096_HPS_FORMAT_FIGEXP M_FIG C_FIG

bioinformatics↗

bioRxiv2021

Simultaneous dimensionality reduction and integration for single-cell ATAC-seq data using deep learning

Advances in single-cell technologies enable the routine interrogation of chromatin accessibility for tens of thousands of single cells, shedding light on gene regulatory processes at an unprecedented resolution. Meanwhile, size, sparsity and high dimensionality of the resulting data continue to pose challenges for its computational analysis, and specifically the integration of data from different sources. We have developed a dedicated computational approach, a variational auto-encoder using a noise model specifically designed for single-cell ATAC-seq data, which facilitates simultaneous dimensionality reduction and batch correction via an adversarial learning strategy. We showcase both its individual advantages on carefully chosen real and simulated data sets, as well as the benefits for detailed cell type characterization via integrating multiple complex datasets.

bioinformatics↗

bioRxiv2021

Herpesviral induction of germline transcription factor DUX4 is critical for viral gene expression

DUX4 is a transcription factor and a master regulator of embryonic genome activation (EGA). During early embryogenesis, EGA is crucial for maternal to zygotic transition at the 8-cell stage in order to overcome silencing of genes and enable transcription from the zygotic genome. In adult somatic cells, DUX4 expression is largely silenced. Activation is likely pathogenic, and in adult muscle cells causes genetic disorder Facioscapulohumeral Muscular Dystrophy (FSHD). We identified activation of DUX4 expression upon lytic replication of the herpesviruses HSV-1, HCMV, EBV and KSHV, but not of adenoviruses, negative strand RNA viruses or positive strand RNA viruses. We demonstrate by RNA-Seq analysis that DUX4 expression upon herpesviral replication leads to the induction of hundreds of DUX4 target genes including germline-specific retroelements as well as several members of the TRIM, PRAMEF and ZSCAN protein families. Moreover, we show that DUX4 expression is a direct consequence of herpesviral infection. DUX4 can be stimulated by overexpression of HSV-1 immediate early proteins, indicating active induction of EGA genes by herpesviral infection. We further show that DUX4 expression is critical for driving HSV-1 gene expression. Our results show that viruses from alpha-, beta- and gamma-herpesvirus subfamilies induce DUX4 expression and downstream germline-specific genes and retroelements. We hypothesize that herpesviruses induce DUX4 expression in order to induce an early embryonic-like transcriptional program that prevents epigenetic silencing of the viral genome and facilitates herpesviral gene expression.

microbiology↗