Speakers
Dr. Nicolas Servant, co-director of the bioinformatics core facility of Institut Curie, Paris, France, will present the lecture "Cancer of Unknown Primary - AI to improve patient care."
Dr. Argyris Papantonis, Professor for Translational Epigenetics at the Medical Faculty of the Georg-August University of Göttingen, Germany, will present the lecture "Repurposed nuclear speckle and mitosis components induce CTCF clustering to sustain the senescence splicing program."
Dr. Carl Herrmann, Professor at the Health Data Science Unit oh the BioQuant Center and Medical Faculty Heidelberg, Germany, will present the lecture "Modelling cellular perturbations using interpretable deep-learning models."
Abstracts
Cancer of Unknown Primary - AI to improve patient care
Cancers of Unknown Primary (CUP) represent 2 to 5 % of all cancers and the 4th cause of cancer mortality wordwide. Among the available therapeutic strategies, molecularly-guided site-specific treatments are a promising approach based on the assumption that CUP are similar in their response to treatment of predicted primary tumours. However, in clinical practice, this strategy only make sense if we are able to efficiently predict the tissue of origin (TOO) of the tumour from the metastatic sample. To address this question, we recently developed a new supervised classification model (transCUPtomics) based on RNA-seq data, which is able to predict the TOO with a high accuracy. We will discuss how this tool has been set up, and how it has been transferred to the French Medicine Genomic plan for use in clinical practice. We will also present new hierarchical machine learning models we are currently testing to improve the model accuracy, and better guide the therapeutic decision.
Repurposed nuclear speckle and mitosis components induce CTCF clustering to sustain the senescence splicing program
Senescence is the end-point of the replicative life span of normal human cells, characterized by a complex sequence of molecular and biochemical events. We recently showed that commitment to the fate of senescence is marked by the dramatic reorganization of CTCF into large senescence-induced CTCF clusters (SICCs). However, the molecular determinants and the functional purpose of SICCs remained unknown. Here, we combine super-resolution imaging, 3D genomics, and functional assays with computational modelling to address both these aspects of SICC emergence. First, we discovered that cells entering senescence repurpose SRRM2 —a key structural component of nuclear speckles— and BANF1 —a protein known to act as ‘molecular glue’ for chromosomes during mitosis— to cluster CTCF on the surface of nuclear speckles. CTCF clusters exhibit properties indicative of phase separation, while rewiring higher-order genome architecture. Second, we could show that CTCF clustering is required for the faithful execution of the senescence splicing program, involving mRNAs encoding key effectors of cell cycle control and chromatin organization. Taken together, this senescence-specific clustering of CTCF on nuclear speckles is a novel paradigm of a change in the biochemical environment of the nucleus being translated into its architectural reorganization to allow deployment of the splicing program underlying cellular ageing.
Modelling cellular perturbations using interpretable deep-learning models
Predicting the effect of cellular perturbations on cells is an important topic with multiple application areas, in particular in cancer research. Perturbations such as drug treatment, genetic alterations, environmental cues, and infection are some examples of such perturbations. Deep-learning models can help to learn perturbation patterns using large-scale single-cell datasets and predict combinatorial effects. In addition, interpretable models can highlight perturbed processes, either regarding signaling or transcriptional regulation. I will review recent development in this field and present our own work in this field, with a focus on tumor cell plasticity and interferon response.