Kelvin Tuong

Developing single-cell analysis methods harnessing adaptive immune receptors

The direct linkage of adaptive immune receptor repertoire with cellular phenotypes from single-cell sequencing technology has proven to be a powerful tool in understanding lymphocyte development and function in healthy and disease contexts. Multi-omics analysis leverages data from different modalities e.g. transcriptomics, epigenetics, proteomics. Recent advances has successfully integrated the data modalities to study cellular biology at an unprecedented resolution. However, unlike the other data modalities, which are largely continuous data, single-cell adaptive immune receptor sequencing (scVDJ-seq) data are a mixture of categorical and continuous data which poses additional challenges for integration. It consists of annotations of variable (V), diversity (D) and joining (J) genes, which are selected and recombined during B/T-cell development. The technology to profile this at the single cell level with paired gene expression data is relatively new. I have led the development of Dandelion, a scVDJ-seq software framework which has been used to deeply profile single-cell lymphocyte biology and I aim to expand on the its capabilities for immunology research.

Developing single-cell analysis methods harnessing adaptive immune receptors

The direct linkage of adaptive immune receptor repertoire with cellular phenotypes from single-cell sequencing technology has proven to be a powerful tool in understanding lymphocyte development and function in healthy and disease contexts. Multi-omics analysis leverages data from different modalities e.g. transcriptomics, epigenetics, proteomics. Recent advances has successfully integrated the data modalities to study cellular biology at an unprecedented resolution. However, unlike the other data modalities, which are largely continuous data, single-cell adaptive immune receptor sequencing (scVDJ-seq) data are a mixture of categorical and continuous data which poses additional challenges for integration. It consists of annotations of variable (V), diversity (D) and joining (J) genes, which are selected and recombined during B/T-cell development. The technology to profile this at the single cell level with paired gene expression data is relatively new. I have led the development of Dandelion, a scVDJ-seq software framework which has been used to deeply profile single-cell lymphocyte biology and I aim to expand on the its capabilities for immunology research.

Building Australia’s Largest Single-Cell Atlas of the Pediatric Immune System

The immune system in early life undergoes rapid and unique developmental changes that are not well captured by adult immune profiling studies. This project aims to generate the most comprehensive single-cell atlas of the pediatric immune system in Australia, based on over 1,000 blood samples from both healthy children and those affected by various diseases, including immune disorders and infections.

We use single-cell RNA sequencing (scRNA-seq) to capture the diversity of immune cells in unprecedented detail, allowing us to map how these cells evolve with age and in response to disease. Importantly, we are integrating scRNA-seq with adaptive immune receptor sequencing (scVDJ-seq) and clinical metadata to uncover how immune repertoires develop and function across different pediatric cohorts.

To analyze this complex dataset, we are developing advanced machine learning and deep learning frameworks capable of identifying age- and disease-specific immune states. We are also applying integrative computational approaches to connect transcriptional states with immune clonotypes and repertoire diversity. This atlas will serve as a key resource for pediatric immunology, providing foundational insights that can inform early diagnostics, vaccine strategies, and targeted therapies for children.

Deep Learning Approaches for Inferring Copy Number Variation from Single-Cell Transcriptomics in Cancer

Copy number variations (CNVs) are a hallmark of cancer, but detecting them accurately at the single-cell level remains a major challenge due to the noisy and sparse nature of single-cell RNA-seq (scRNA-seq) data. This project focuses on the development of deep learning models capable of inferring CNVs directly from transcriptomic profiles, with the ultimate goal of identifying cancer cells within heterogeneous tissues.

By applying neural network architectures tailored to single-cell data, we aim to model the underlying gene dosage effects reflected in gene expression patterns. Our models will be trained to distinguish malignant from normal cells and to identify patterns of genomic instability across tumor subclones. These methods will allow us to reconstruct the evolutionary dynamics of tumors and better understand the relationship between transcriptomic signatures and chromosomal alterations.

This work has direct applications in cancer diagnostics, especially in improving the resolution of tumor profiling for early detection and minimal residual disease monitoring. We also aim to make these methods publicly available as open-source tools to support the broader community studying tumor heterogeneity through single-cell approaches.