Exploring the heterogeneity of cellular senescence through integrative analysis of bio-image and -omic data
Ebony Watson and Jessica Mar
The Australian Institute for Bioengineering and Nanotechnology, University of Queensland, Queensland, Australia
Advances in imaging technologies have enabled the high-throughput visualisation of biological phenotypes, from the molecular to whole-organism scale. Simultaneously, new computational techniques have emerged, facilitating the precise and objective quantification of such images. These developments present a promising framework, whereby features from images can be integrated with -omics data to explore the molecular drivers of biological phenotypes in innovative and exciting ways.
The power of multimodal analysis is particularly applicable to biological ageing, where our understanding of regulatory mechanisms is impeded by the complexity and heterogeneity of many age-associated phenotypes, such as senescence. Senescence is a state of irreversible cell-cycle arrest, induced in response to stressors such as telomere shortening or specific mitogenic signals. This mechanism exists to prevent the proliferation of potentially damaged DNA, however is accompanied by a Senescence Associated Secretory Phenotype, comprised of numerous pro-inflammatory molecules. As the number of senescent cells in a tissue increases, their collective secretions wreak havoc on neighbouring cells, resulting in the many diseases commonly associated with ageing. Clearance of senescent cells in model organisms can reduce or cure these age-associated pathologies, and is therefore an area of considerable interest for the development of clinical treatments.
However despite the ubiquity of senescence throughout the human body, and ageing disorders, it is an extremely dynamic and heterogeneous state. Bio-markers of senescence vary significantly with cell-type, inducing-stressor, and temporally. Here we present the development of a multimodal deep neural network to appropriately integrate several image and -omic data types, each characterising a distinct phenotype from senescent cells induced by different stressors. This model, along with subsequent integrative statistical analyses, will enable the deconvolution of multiple senescence-associated genotype-phenotype interactions. Through predictive classification, the model will also assist in the definition of distinct senescence sub-types, enabling the targeting specificity required to develop effective treatments for senescence.