Phasing of Partially Resolved Metagenomic Assemblies and Identifying new CAR-T targets from single cell RNA-seq data

About this Class

For our next CDSL webinar we will have presentations by two CDSL fellows: Ekaterina Kazantseva and Sanna Madan.  Ekaterina is a master’s student in Dr. Mikhail Kolmogorov's group and the title of her talk is "Phasing of Partially Resolved Metagenomic Assemblies". Sanna is a PhD student working with Dr. Eytan Ruppin and she will give a talk on "Identifying new CAR-T targets from single cell RNA-seq data". Abstract (Ekaterina's talk): Long-read metagenomic sequencing has recently been used to recover complete bacterial genomes from various complex metagenomic communities. Metagenome assembly algorithms however are still facing challenges in deconvolution of closely-related species and strains. De novo assemblies of highly heterogeneous bacterial species typically result in tangled assembly graphs, where some sequences could be strain-specific, while others represent species-level consensus. Such a partially-collapsed representation of bacterial strains does not take full advantage of the ability of long reads to phase small variants. In this work we present an algorithm called MetaPhase that extends metagenomic phasing approaches to assembly graphs. Our algorithm operates on graph paths rather than single contigs, and iteratively simplifies assembly graphs with newly reconstructed strain contigs. We benchmark our algorithm using mock communities and show that it produces accurate and complete strain-level reconstructions and substantially improves over the initial partially-collapsed assemblies. Abstract (Sanna's talk): Chimeric antigen receptor (CAR) T cell therapy is a powerful and promising tool for unleashing lasting antitumor immunity. While this modality has yielded major clinical success in treating blood cancers, obstacles remain to achieve its potential in solid tumors. In particular, identifying targets that are uniformly expressed across cancer cells and minimally so on normal cells remains a key challenge. Thus, the criteria for ideal CAR-T targets are two-fold: they must (1) be selectively expressed in tumor cells and not on non-tumor cells within the tumor microenvironment (TME), and (2) be lowly expressed across normal human tissues. Mining single cell transcriptomics datasets of solid tumors, we first survey the landscape of current CAR-T targets in the clinic, charting their tumor cell-specificity (termed selectivity score) and expression levels across healthy tissues (termed safety score). Next, we identify cell surface protein-encoding genes whose selectivity and safety scores surpass those of the leading targets in clinics. Subsequently, we put forth that the proteins encoding the genes resulting from our analysis may constitute optimal new targets of CAR-T therapies. Intriguingly, our analysis has yielded an enrichment of targets for head and neck cancer, a cancer type for which there are currently very few unique targets of CAR-T therapies in the clinic. Taken together, this analysis uncovers a large potential of scRNA-seq data in developing precise, selective CAR-T therapies.