85 research outputs found
Visualization and analysis strategies for dynamic gene-phenotype relationships and their biological interpretation
No full textThe complexity of biological systems is one of their most fascinating and, at the same time, most cryptic aspects. Despite the progress of technology that has enabled measuring biological parameters at deeper levels of detail in time and space, the ability to decipher meaning from these large amounts of heterogeneous data is limited. In order to address this challenge, both analysis and visualization strategies need to be adapted to handle this complexity.
At system-wide level, we are still limited in our ability to infer genetic and
environmental causes of disease, or consistently compare and link phenotypes. Moreover, despite the increasing availability of time-resolved experiments, the temporal context is often lost. In my thesis, I explored a series of analysis and visualization strategies to compare and connect dynamic phenotypic outcomes of cellular perturbations in a genetic and network context.
More specifically, in the first part of my thesis, I focused on the cell cycle as one of the best examples of a complex, highly dynamic process. I applied analysis and data integration methods to investigate phenotypes derived from cell division failure. I examined how such phenotypes may arise as a result of perturbations in the underlying network. To this purpose, I investigated the role of short structural elements at binding interfaces of proteins, called linear motifs, in shaping the cell division network. I assessed their association to different phenotypes, in the context of local perturbations and of disease.
This analysis enabled a more detailed understanding of the regulatory mechanisms beyond the malfunctioning of cell division processes, but the ability to compare phenotypes and track their evolution was limited. Exploring large-scale, time-resolved phenotypic screens is still a bottleneck, especially in the visualization area. To help address this question, in the subsequent parts of the thesis I proposed novel visualization approaches that would leverage pattern discovery in such heterogeneous, dynamic datasets and enable the generation of new hypotheses.
First, I extended an existing visualization tool, Arena3D, to enable the comparison of phenotypes in a genetic and network context. I used this tool to continue the exploration of phenotype-wide differences between outcomes of gene function suppression within mitosis. I also applied it to an investigation of systemic changes in the network of embryonic stem cell fate determinants upon downregulation of the pluripotency factor Nanog.
Second, time-resolved tracking of phenotypes opens up new possibilities in exploring how genetic and phenotypic connections evolve through time, an aspect that is largely missing in the visualization area. I developed a novel visualization approach that uses 2D/3D projections to enable the discovery of genetic determinants linking phenotypes through time. I used the resulting tool, PhenoTimer, to investigate the patterns of transitions between phenotypes in cell populations upon perturbation of cell division and the timing of cancer-relevant transcriptional events. I showed the potential of discovering drug synergistic effects by visual mapping of similarities in their mechanisms of action. Overall, these approaches help clarify aspects of the consequences of cell division failure and provide general visualization frameworks that should be of interest to the wider scientific community, for use in the analysis of multidimensional phenotypic screens
SpottedPy quantifies relationships between spatial transcriptomic hotspots and uncovers environmental cues of epithelial-mesenchymal plasticity in breast cancer
No full textAbstract Spatial transcriptomics is revolutionizing the exploration of intratissue heterogeneity in cancer, yet capturing cellular niches and their spatial relationships remains challenging. We introduce SpottedPy, a Python package designed to identify tumor hotspots and map spatial interactions within the cancer ecosystem. Using SpottedPy, we examine epithelial-mesenchymal plasticity in breast cancer and highlight stable niches associated with angiogenic and hypoxic regions, shielded by CAFs and macrophages. Hybrid and mesenchymal hotspot distribution follows transformation gradients reflecting progressive immunosuppression. Our method offers flexibility to explore spatial relationships at different scales, from immediate neighbors to broader tissue modules, providing new insights into tumor microenvironment dynamics
Visualizing time-related data in biology, a review
No full textpeer reviewedTime is of the essence, also in biology. Monitoring disease progression or timing developmental defects are key aspects in the process of drug discovery and therapy trial. Furthermore, before deciphering the course of evolution of these complex processes, we need an understanding of the basic dynamics of biological phenomena that are often strictly time-regulated (e.g. circadian rhythms). With the advances in technologies able to measure timing effects and dynamics of regulatory aspects, visualization and analysis tools try to keep up the pace with the new challenge. Beyond the classical timeline plots, notable attempts at more involved temporal interpretation have been made in the recent years, but awareness of the available resources is still limited within the scientific community. Here we review some of the advances in biological visualization of time-driven processes and look at how they allow analyzing data now and in the future
PhenoTimer: Software for the Visual Mapping of Time-Resolved Phenotypic Landscapes
No full textpeer reviewedTiming common and specific modulators of disease progression is crucial for treatment, but the understanding of the underlying complex system of interactions is limited. While attempts at elucidating this experimentally have produced enormous amounts of phenotypic data, tools that are able to visualize and analyze them are scarce and the insight obtained from the data is often unsatisfactory. Linking and visualizing processes from genes to phenotypes and back,
in a temporal context, remains a challenge in systems biology. We introduce PhenoTimer, a 2D/3D visualization tool for the mapping of time-resolved phenotypic links in a genetic context. It uses a novel visualization approach for relations between morphological defects, pathways or diseases, to enable fast pattern discovery and hypothesis generation. We illustrate its capabilities of tracing dynamic motifs on cell cycle datasets that explore the phenotypic
order of events upon perturbations of the system, transcriptional activity programs and their connection to disease. By using this tool we are able to fine-grain regulatory programs for individual time points of the cell cycle and better understand which patterns arise when these programs fail. We also illustrate a way to identify common mechanisms of misregulation in diseases and drug abuse
Similar mechanisms of drug-induced gene regulation for up to 8 hours after treatment.
No full text<p>(A) Pair-wise connections between drugs and networks of genes affected by the same drug are visualized using PhenoTimer for every time point. Two drugs are connected if they similarly regulate the same gene. Two genes are connected in the network if they respond to the same drug(s). Thickness of links corresponds to the number of genes, or drugs, respectively, shared by two partners. Only genes with transcription values in the lower quartile (top) or upper quartile (bottom) are taken into account. The core gene network is depicted in yellow and the variable gene elements (i.e. that don’t appear at every time point) are highlighted in orange (lower quartile) and green (upper quartile). Links between drugs are depicted in magenta and pink if they contain the highest number of commonly regulated genes. The genes specific for that particular link only are circled in the same color in the network below. (B) Heat map of the gene expression values after 8 hours of drug induction is shown for every drug. The line corresponding to gene <i>Tnfrsf25</i> is highlighted and the columns corresponding to ethanol and heroin are also indicated. The graphic of transcription counts throughout the time course for this gene after heroin induction is shown below. Both images have been generated using PhenoTimer. (C) The network of human homologs for the relatively lowly and highly expressed genes. The variable genes are highlighted in orange (lower quartile) and green (upper quartile), along with the time points when their values are in the required quartile range.</p
Genomic Analysis of Response to Neoadjuvant Chemotherapy in Esophageal Adenocarcinoma.
No full textNeoadjuvant therapy followed by surgery is the standard of care for locally advanced esophageal adenocarcinoma (EAC). Unfortunately, response to neoadjuvant chemotherapy (NAC) is poor (20-37%), as is the overall survival benefit at five years (9%). The EAC genome is complex and heterogeneous between patients, and it is not yet understood whether specific mutational patterns may result in chemotherapy sensitivity or resistance. To identify associations between genomic events and response to NAC in EAC, a comparative genomic analysis was performed in 65 patients with extensive clinical and pathological annotation using whole-genome sequencing (WGS). We defined response using Mandard Tumor Regression Grade (TRG), with responders classified as TRG1-2 (n = 27) and non-responders classified as TRG4-5 (n =38). We report a higher non-synonymous mutation burden in responders (median 2.08/Mb vs. 1.70/Mb, p = 0.036) and elevated copy number variation in non-responders (282 vs. 136/patient, p < 0.001). We identified copy number variants unique to each group in our cohort, with cell cycle (CDKN2A, CCND1), c-Myc (MYC), RTK/PIK3 (KRAS, EGFR) and gastrointestinal differentiation (GATA6) pathway genes being specifically altered in non-responders. Of note, NAV3 mutations were exclusively present in the non-responder group with a frequency of 22%. Thus, lower mutation burden, higher chromosomal instability and specific copy number alterations are associated with resistance to NAC
Teaching Pathology Foundation Models to Accurately Predict Gene Expression with Parameter Efficient Knowledge Transfer
No full textGene expression profiling provides critical insights into cellular heterogeneity, biological processes, and disease mechanisms. There has been an increasing interest in computational approaches that can predict gene expression directly from digitalized histopathology images. While image foundation models have shown promise in a variety of pathology downstream analysis, their performances on gene expression prediction are still limited. Explicitly incorporating information from the transcriptomic models can help image models address domain shift, yet the fine-tuning and alignment of foundation models can be expensive. In this work, we propose Parameter Efficient Knowledge trAnsfer (PEKA), a novel framework that leverages Block-Affine Adaptation and integrates knowledge distillation and structure alignment losses for cross-modal knowledge transfer. We evaluated PEKA for gene expression prediction using multiple spatial transcriptomics datasets (comprising 206,123 image tiles with matched gene expression profiles) that included various types of tissue. PEKA achieved at least 5% performance improvement over baseline foundation models while also outperforming alternative parameter-efficient fine-tuning strategies. We have released the code, datasets and aligned models at Github to facilitate broader adoption and further development for parameter efficient model alignment
Phenotypic transition patterns in cell populations upon knockdown of genes essential for cell division.
No full text<p>Three arc representations are shown: (A) 3D; (B) linear 2D; (C) circular 2D. An arc represents a transition from one phenotype to the other at consecutive time points. The color of the arc corresponds to the phenotype into which the cells transition. The height (3D) and width (2D), respectively, of arcs indicates the number of genes whose suppression causes the respective phenotypic transition at that particular moment (at most 5 genes for this dataset). The GO term network in the boxed picture in the upper right corner highlights (in red) the molecular functions of the genes whose knockdown causes a transition at time point 41. The respective transitions are shown as arcs in the plot for the particular time point. The size of the nodes in the network is proportional to the number of genes in the dataset that are enriched for the respective function. The GO network was generated using BiNGO [37] and subsequently loaded into PhenoTimer.</p
Cell environment shapes TDP-43 function with implications in neuronal and muscle disease
No full textTDP-43 (TAR DNA-binding protein 43) aggregation and redistribution are recognised as a hallmark of amyotrophic lateral sclerosis and frontotemporal dementia. As TDP-43 inclusions have recently been described in the muscle of inclusion body myositis patients, this highlights the need to understand the role of TDP-43 beyond the central nervous system. Using RNA-seq, we directly compare TDP-43-mediated RNA processing in muscle (C2C12) and neuronal (NSC34) mouse cells. TDP-43 displays a cell-type-characteristic behaviour targeting unique transcripts in each cell-type, which is due to characteristic expression of RNA-binding proteins, that influence TDP-43's performance and define cell-type specific splicing. Among splicing events commonly dysregulated in both cell lines, we identify some that are TDP-43-dependent also in human cells. Inclusion levels of these alternative exons are altered in tissues of patients suffering from FTLD and IBM. We therefore propose that TDP-43 dysfunction contributes to disease development either in a common or a tissue-specific manner
Arena3D: visualizing time-driven phenotypic differences in biological systems
No full textAbstract Background Elucidating the genotype-phenotype connection is one of the big challenges of modern molecular biology. To fully understand this connection, it is necessary to consider the underlying networks and the time factor. In this context of data deluge and heterogeneous information, visualization plays an essential role in interpreting complex and dynamic topologies. Thus, software that is able to bring the network, phenotypic and temporal information together is needed. Arena3D has been previously introduced as a tool that facilitates link discovery between processes. It uses a layered display to separate different levels of information while emphasizing the connections between them. We present novel developments of the tool for the visualization and analysis of dynamic genotype-phenotype landscapes. Results Version 2.0 introduces novel features that allow handling time course data in a phenotypic context. Gene expression levels or other measures can be loaded and visualized at different time points and phenotypic comparison is facilitated through clustering and correlation display or highlighting of impacting changes through time. Similarity scoring allows the identification of global patterns in dynamic heterogeneous data. In this paper we demonstrate the utility of the tool on two distinct biological problems of different scales. First, we analyze a medium scale dataset that looks at perturbation effects of the pluripotency regulator Nanog in murine embryonic stem cells. Dynamic cluster analysis suggests alternative indirect links between Nanog and other proteins in the core stem cell network. Moreover, recurrent correlations from the epigenetic to the translational level are identified. Second, we investigate a large scale dataset consisting of genome-wide knockdown screens for human genes essential in the mitotic process. Here, a potential new role for the gene lsm14a in cytokinesis is suggested. We also show how phenotypic patterning allows for extensive comparison and identification of high impact knockdown targets. Conclusions We present a new visualization approach for perturbation screens with multiple phenotypic outcomes. The novel functionality implemented in Arena3D enables effective understanding and comparison of temporal patterns within morphological layers, to help with the system-wide analysis of dynamic processes. Arena3D is available free of charge for academics as a downloadable standalone application from: http://arena3d.org/.</p
- …
