Findings on the development and use of the AI tool, called AAnet, have today been published in Cancer Discovery, a journal of the American Association for Cancer Research.
Tumours aren't made up of just one cell type - they're a mix of different cells that grow and respond to treatment in different ways. This diversity, or heterogeneity, makes cancer harder to treat and can in turn lead to worse outcomes, especially in triple-negative breast cancer.
"Heterogeneity is a problem because currently we treat tumours as if they are made up of the same cell. This means we give one therapy that kills most cells in the tumour by targeting a particular mechanism. But not all cancer cells may share that mechanism. As a result, while the patient may have an initial response, the remaining cells can grow and the cancer may come back," says Associate Professor Christine Chaffer, co-senior author of the study and Co-Director of the Cancer Plasticity and Dormancy Program at Garvan.
But while heterogeneity is a problem, researchers don’t know enough to characterise it: "So far researchers haven’t been able to clearly explain how adjacent cells in a tumour differ from each other, and how to classify those differences into meaningful ways to better treat tumours. But this is exactly what we need to know so we can kill all cells within that tumour with the right therapies," Associate Professor Chaffer adds.
To solve this problem, the team developed and trained a powerful new AI tool called AAnet that can detect biological patterns in cells within tumours.
They then used the AI tool to uncover patterns in the level of gene expression of individual cells within tumours, focusing on preclinical models of triple-negative breast cancer and human samples of ER positive, HER2 positive and triple-negative breast cancer. Through this, they identified five different cancer cell groups within a tumour, with distinct gene expression profiles that indicated vast differences in cell behaviour.
"By using our AI tool, we were consistently able to discover five new groups of cell types within single tumours called ‘archetypes’. Each group exhibited different biological pathways and propensities for growth, metastasis and markers of poor prognosis. Our next steps are to see how these groups may change over time, for example before and after chemotherapy," says Associate Professor Chaffer.
This is a first for cancer research. Co-lead, Associate Professor Smita Krishnaswamy from Yale University who led the development of the AI tool states: "Thanks to technology advances, the last 20 years have seen an explosion of data at the single-cell level. With this data we have been finding out that not only is each patient’s cancer different, but each cancer cell behaves differently from another. Our study is the first time that single-cell data have been able to simplify this continuum of cell states into a handful of meaningful archetypes through which diversity can be analysed to find meaningful associations with spatial tumour growth and metabolomic signatures. This could be a game changer."
The researchers say the use of AAnet to characterise the different groups of cells in a tumour according to their biology opens doors for a paradigm shift in how we treat cancer.
"Currently the choice of cancer treatment for a patient is largely based on the organ that the cancer came from such as breast, lung or prostate and any molecular markers it may exhibit. But this assumes that all cells in that cancer are the same. Instead, now we have a tool to characterise the heterogeneity of a patient’s tumour and really understand what each group of cells is doing at a biological level. With AAnet, we now hope to improve the rational design of combination therapies that we know will target each of those different groups through their biological pathways. This has the potential to vastly improve outcomes for that patient," says Associate Professor Chaffer.
On the application of AAnet, co-senior author of the study and Chief Scientific Officer of Garvan Professor Sarah Kummerfeld states: "We envision a future where doctors combine this AI analysis with traditional cancer diagnoses to develop more personalised treatments that target all cell types within a person’s unique tumour. These results represent a true melding of cutting-edge technology and biology that can improve patient care. Our study focused on breast cancer, but it could be applied to other cancers and illnesses such as autoimmune disorders. The technology is already there."
Venkat A, Youlten SE, San Juan BP, Purcell CA, Gupta S, Amodio M, Neumann DP, Lock JG, Westacott AE, McCool CS, Burkhardt DB, Benz A, Mollbrink A, Lundeberg J, van Dijk D, Holst J, Goldstein LD, Kummerfeld S, Krishnaswamy S, Chaffer CL.
AAnet resolves a continuum of spatially-localized cell states to unveil intratumoral heterogeneity.
Cancer Discov. 2025 Jun 24. doi: 10.1158/2159-8290.CD-24-0684