Dr Stefano Mangiola

The Computational Cancer Immunogenomics group, led by Dr. Mangiola, is interested in the immune system's role in cancer progression and treatment response studies through high-throughput data modelling. 

By profiling a patient’s immune system through modern spatial and single-cell technologies, we model the propensity to enter metastatic progression and be resilient to metastatic spread (e.g., in breast cancer). Similarly, we intend to identify systemic immune features that explain local immunity (within the tumour microenvironment) and predict resistance to neoadjuvant therapy in breast and other cancer types.

The immune system is diverse across the human population. We pioneered population-scale immune system modelling using large-scale single-cell data (Human Cell Atlas) and quantified its heterogeneity across tissues. This heterogeneity includes tissue-specific ageing programs, sexual dimorphism, and ethnical diversity in immunotherapy targets. Now, we aim to use artificial intelligence (AI) models (i.e. LLM) to extend our immune map to cancer. Specifically, we are interested in building foundation models that can identify stable immunotherapy targets across ethnic groups.

Our work includes the construction of scalable infrastructure and interfaces that allow multi-atlas-level analyses and annotation. This includes tidyomics, CuratedAtlasQuery and HPCell.

 We are particularly interested in the following areas:

1) Integration of spatial and single-cell transcriptomics and proteomics.

2) Machine learning and classification.

3) Large-language AI models applied to cellular biology.

4) Cancer immunodiagnosis.

5) R tidy programming applied to multiomics.

6) Large-scale inference from single-cell multi-atlases.

Higher Degree Research Projects

1) Model the immune system with large-language AI models. 

Demographic factors like age and sex critically influence disease outcomes and treatment responses, including cancer treatments targeting the immune system (immunotherapy). Australia has an ethnically diverse and ageing population. Despite its importance, a unified resource to test the population-level diversity of therapy and diagnosis targets is missing, with the current paradigm of population-level investigation of complex cancer traits still relying on homogenised tissue data such as 15-year-old TCGA. With a recent breakthrough3, we pioneered the population-level immune system investigation from massive-scale single-cell resources. Across 12,981 healthy individuals and 30 organs, we mapped profound immunological changes across ageing, sexes and ethnicity, uncovering diversity in key pathways used in immunotherapy (e.g. LAG3, SLAMF7 and CD83). This project aims to quantify the population-level diversity of immunotherapy targets, providing the scientific and clinical community with an unprecedented encyclopaedic resource. We will translate our infrastructure to generative single-cell large-language models (ChatGPT-like) to identify novel immune targets conserved in ageing and across sexes. These artificial intelligence (AI) models are revolutionising data-driven cell biology research. However, they were not designed for clinically related questions. We will pivot this technology to our massive clinically annotated cancer cell compendium. We will then improve the current static-cell paradigm by extending AI models to a dynamic representation of cells. 

2) Integrate spatial and single-cell multiomics to predict neoadjuvant response in cancer with locally-assisted systemic immunodiagnosis.

The effective prediction of neoadjuvant therapy outcomes in cancer treatment remains a pivotal challenge. This project proposes a novel approach by integrating spatial and single-cell multi-omics analyses to enhance the precision of neoadjuvant response predictions. Our methodology leverages cutting-edge techniques in both spatial omics and single-cell sequencing to capture a comprehensive molecular landscape at the tumour site. By delineating the intricate interplay between tumour cells and the immune environment, this approach aims to uncover specific biomarkers and signalling pathways indicative of therapy responses.

Further innovation lies in implementing locally-assisted systemic immunodiagnosis, which utilises local tumour data to inform systemic immune profiling. This dual approach promises to improve the accuracy of predicting patient responses to neoadjuvant therapies and aims to personalise treatment plans, thereby potentially enhancing clinical outcomes.

The project's multi-disciplinary framework combines advanced bioinformatics tools with clinical oncology insights, enabling a more targeted and efficient diagnostic process. By predicting therapeutic efficacy before treatment initiation, this strategy seeks to spare patients from the adverse effects of ineffective therapies and streamline clinical decision-making. 

3) Build large-scale and scalable infrastructure for analysis, deployment and exploration of the single-cell universe.

Single-cell and spatial omic technologies have fundamentally transformed biological research by generating vast quantities of data. This influx challenges existing bioinformatics pipelines and the capacity of individual users to keep up with the rapidly evolving demands of impactful data-driven research. In collaboration with CSL, this project enhances the capabilities of CuratedAtlasQuery and HPCell to establish a privately deployable intelligence hub for single-cell and spatial data. CuratedAtlasQuery has already facilitated extensive profiling of the immune system at the human population level. We aim to expand this database by integrating biological annotations and data summarisation techniques to democratise access to large-scale single-cell analyses. HPCell is being developed as an analytical language that allows the execution of massively parallel single-cell analysis workflows in a tidy R style, and enables their deployment on high-performance computing platforms.

4) Tidyomics

Tidyomics (Nat Methods., 2024) is an R software ecosystem that enhances the analysis and visualisation of high-dimensional omics data, applying the principles of tidy data analysis, a de facto standard in data science. Given its extensive adoption, we propose to improve the documentation, robustness, and interoperability of the Tidyomics ecosystem and extend it to spatial profiling technologies. Tidyomics packages enable computational biologists to employ a user-friendly grammar to manipulate popular data containers across omics (genomics, transcriptomics, cytometry) and platforms (Bioconductor, Seurat). Tidyomics aggregates a growing user base and community of developers, forming an international network that spans five continents. In a manuscript (Hutchison and Keyes 2023), we formalised the Tidyomics ecosystem and established a roadmap with our community GitHub Project space. Single-cell data repositories like The Human Cell Atlas and Curated Cancer Atlas drive next-generation research. Examining tissue biology at the single-cell level is refining our choice of cell and gene markers for specific groups, organs, and cells. We are focusing on several enhancements to the Tidyomics ecosystem for better interfacing with large-scale single-cell atlas collections.