AI breakthrough reveals hidden cancer cell groups, paving way for more targeted treatments

AI tool set to transform characterisation and treatment of cancers

Scientists have developed an AI tool that can enhance cancer diagnostics and guide more effective treatment strategies.

Not all tumour cells are the same

Tumours consist of diverse cell populations, each with unique growth patterns and responses to treatment. This heterogeneity complicates therapy, often leading to poorer outcomes, particularly in triple-negative breast cancer.

Currently, treatments often target a single mechanism assuming uniformity across cancer cells. However, not all cells share the same vulnerabilities, which can result in some cells surviving therapy and causing relapse.

Despite its significance, tumour heterogeneity remains poorly characterised. Researchers have struggled to classify how adjacent cells differ within a tumour in ways that meaningfully inform treatment choices.

A new tool characterises five distinct cancer cell groups

To address this, researchers created an AI tool named AAnet that detects biological patterns in tumour cells by analysing gene expression at the single-cell level. The tool was applied to preclinical models and human samples from various breast cancer subtypes, including triple-negative, ER positive, and HER2 positive cancers.

AAnet consistently identified five distinct cancer cell groups within tumours, each exhibiting unique gene expression profiles. These profiles correspond to differences in growth rates, metastatic potential, and prognosis markers.

Future research will explore how these cell groups evolve, especially in response to chemotherapy, to better understand tumour dynamics over time.

New classification to drive better, targeted treatments

Using AAnet to characterise tumour cell diversity offers a new framework for cancer treatment. Instead of basing therapy mainly on the tumour’s tissue of origin and a few molecular markers, this approach focuses on the biological behaviour of distinct cell groups within a tumour.

This enables the rational design of combination therapies aimed at targeting each identified cell group through their specific biological pathways, potentially improving patient outcomes significantly.

Looking ahead, integrating AI analysis like AAnet with standard diagnostic methods could allow clinicians to develop personalized treatments that effectively target all tumour cell types within a patient’s cancer. While this study focused on breast cancer, the approach may be applicable to other cancers and diseases such as autoimmune disorders.

This research was supported by institutions in Australia and the US.

Key Researchers

• Associate Professor Christine Chaffer – Co-Director of the Cancer Plasticity and Dormancy Program, Garvan Institute of Medical Research
• Associate Professor Smita Krishnaswamy – Genetics and Computer Science, Yale School of Medicine, lead developer of AAnet
• Professor Sarah Kummerfeld – Chief Scientific Officer, Garvan Institute of Medical Research

Published in: Cancer Discovery

Article title: AAnet resolves a continuum of spatially-localized cell states to unveil intratumoral heterogeneity

Publication date: 24-Jun-2025