Biology subjects

Anselmi, F.

Publications and source records attributed to Anselmi, F..

3 recordsLinked to original sources

bioRxiv2023

A Bayesian method to infer copy number clones from single-cell RNA and ATAC sequencing

Single-cell RNA and ATAC sequencing technologies allow one to probe expression and chromatin accessibility states as a proxy for cellular phenotypes at the resolution of individual cells. A key challenge of cancer research is to consistently map such states on genetic clones, within an evolutionary framework. To this end we introduce CONGAS+, a Bayesian model to map single-cell RNA and ATAC profiles generated from independent or multimodal assays on the latent space of copy numbers clones. CONGAS+ can detect tumour subclones associated with aneuploidy by clustering cells with the same ploidy profile. The framework is implemented in a probabilistic language that can scale to analyse thousands of cells thanks to GPU deployment. Our tool exhibits robust performance on simulations and real data, highlighting the advantage of detecting aneuploidy from two distinct molecules as opposed to other single-molecule models, and also leveraging real multi-omic data. In the application to prostate cancer, lymphoma and basal cell carcinoma, CONGAS+ did retrieve complex subclonal architectures while providing a coherent mapping among ATAC and RNA, facilitating the study of genotype-phenotype mapping, and their relation to tumour aneuploidy. Author summaryAneuploidy is a condition caused by copy number alterations (CNAs), which brings cells to acquire or lose chromosomes. In the context of cancer progression and treatment response, aneuploidy is a key factor driving cancer clonal dynamics, and measuring CNAs from modern sequencing assays is therefore important. In this framing, we approach this problem from new single-cell assays that measure both chromatin accessibility and RNA transcripts. We model the relation between single-cell data and CNAs and, thanks to a sophisticated Bayesian model, we are capable of determining tumour clones from clusters of cells with the same copy numbers. Our model works when input cells are sequenced independently for both assays, or even when modern multi-omics protocols are used. By linking aneuploidy to gene expression and chromatin conformation, our new approach provides a novel way to map complex genotypes with phenotype-level information, one of the missing factors to understand the molecular basis of cancer heterogeneity.

bioinformatics↗

bioRxiv2023

Lifelong tissue memory relies on spatially organised dedicated progenitors located distally from the injury

It is believed epithelial cells that have participated in a wound repair elicit a more efficient but locally restricted response to future injuries. However here we show that the cell adaptation resulting from a localised tissue damage has a wide spatial impact at a scale not previously noticed. We demonstrate that away from injured site, after a first injury a specific epithelial stem cell population gives rise to long term wound-memory progenitors residing in their own niche of origin. Notably these progenitors have not taken part in the first wound healing but become pre-activated through priming. This adaptation differs from classical features of trained immunity previously shown to be adopted by other epithelial stem cells. Our newly identified wound-distal memory cells display a cell-autonomous transcriptional pre-activated state leading to an enhanced wound repair ability that can be partially recapitulated through epigenetic perturbation even in absence of an injury. Importantly, the harmful consequences of wound repair, such as exacerbated tumorigenesis, occur within these primed cells and follow their spatial distribution. Overall, we show that sub-organ scale adaptation of an injury relies on spatially organised and memory-dedicated progenitors, characterised by an epigenetic actionable cell state, that predisposes to tumour onset.

cell biology↗

bioRxiv2022

Robust deep learning object recognition models rely on low frequency information in natural images

Machine learning models have difficulty generalizing to data outside of the distribution they were trained on. In particular, vision models are usually vulnerable to adversarial attacks or common corruptions, to which the human visual system is robust. Recent studies have found that regularizing machine learning models to favor brain-like representations can improve model robustness, but it is unclear why. We hypothesize that the increased model robustness is partly due to the low spatial frequency preference inherited from the neural representation. We tested this simple hypothesis with several frequency-oriented analyses, including the design and use of hybrid images to probe model frequency sensitivity directly. We also examined many other publicly available robust models that were trained on adversarial images or with data augmentation, and found that all these robust models showed a greater preference to low spatial frequency information. We show that preprocessing by blurring can serve as a defense mechanism against both adversarial attacks and common corruptions, further confirming our hypothesis and demonstrating the utility of low spatial frequency information in robust object recognition.

neuroscience↗