Dr. Theodore Perkins shares insights from an ongoing study on T-cell Acute Lymphoblastic Leukaemia (T-ALL), focusing on improving detection of copy number variations (CNVs) using multi-omic single-cell data. The goal is to identify elusive cancer stem cells that are often responsible for relapse after chemotherapy, despite treatment success in many patients. 

To understand the molecular underpinnings of these stem-like cells, Perkins' team employed TEA-seq, a single-cell tri-modal profiling method that simultaneously measures RNA expression, chromatin accessibility, and surface proteins. Eight patient samples were collected, including matched diagnostic and relapse samples. Special care was taken to distinguish tumour cells from normal cells using multiple data modalities and prior sequencing data. 

The presentation focuses on the difficulty of detecting CNVs using TEA-seq data. Traditional DNA-based methods rely on read density variations, but in single-cell transcriptomic or epigenomic data, regulatory noise and expression variability often obscure true CNVs. Several computational tools exist to infer CNVs from RNA or ATAC-seq data, but recent benchmarks suggest only partial success and inconsistent results. 

To address this, Perkins' team applied a simplified, custom method, histogramming read density per chromosome across cells. This revealed clear CNV patterns, including a trimodal distribution on chromosome 4 in relapse samples, corresponding to known subclones. Aggregate read-depth comparisons between tumour and normal cells also uncovered complete deletions on chromosomes 9 and 14—signals missed by tools like EpiAneufinder, which discard low-coverage regions. 

The study highlights the limitations of existing CNV detection tools and emphasises the value of simple, transparent analyses. Perkins concludes that better tumour/normal cell classification and more intuitive visualizations may outperform current black-box algorithms. His lab is currently developing a software for CNV detection, available for early use and feedback. 

This work underscores the importance of robust, interpretable tools for leveraging multi-omic data in cancer research and personalised medicine.