NCIG researchers announced as Google.org Impact Challenge: AI for Science recipients
Researchers from the National Centre for Indigenous Genomics have been announced as recipients of the Google.org Impact Challenge: AI for Science.
NCIG is one of 17 worldwide research teams receiving a total of USD$36 million through the Google.org Impact Challenge: AI for Science to advance innovative AI-powered solutions to global challenges. Further information and details of other recipients can be found here.
Read on further for a summary of the project.
An atlas of chemotherapy resistance, built to read a tumour's resistance programme from its DNA
The Cytotoxic Resistance Atlas for AI-enabled Precision Oncology is a three-year project led through the John Curtin School of Medical Research, Australian National University in a collaborative effort with the University of Kent, PacBio, and the Victorian Cancer Biobank. It will measure how chemotherapy-resistant cancer cells differ from sensitive ones and train AI models to recognise the same resistance programmes in patient tumours. The data, models and software will be released openly.
The problem
Resistance to chemotherapy is a primary barrier to effective cancer care. It is rarely the work of a single gene, generally arising from coordinated changes across many genes, proteins, and regulatory pathways, which single-gene marker tests do not capture. Without a way to read these multi-gene programmes, the choice of chemotherapy often proceeds by trial and error, exposing patients to drugs that will not work and to their side effects.
The approach
The atlas will profile cancer cell lines, some resistant to chemotherapy and some sensitive, each measured four ways: gene activity in single cells, whole genomes, full-length RNA, and proteins. From this atlas, two AI models will be built. The first will identify the multi-gene resistance programmes active in a tumour. The second will infer those programmes from tumour DNA alone, where no RNA or protein data exist. Predictions will be tested by gene knockout in the laboratory and against patient tumour data.
Figure 1 | The problem, and the plan from laboratory to clinic. (a) Each square is a map of how genes interact, one row and one column per gene. A single-gene test reads one marker (gold row). Resistance usually runs through groups of genes acting together, shown as coloured blocks along the diagonal. (b) Top lane - laboratory: cancer cell lines, sensitive and resistant, are each measured four ways to build the open Resistance Atlas, which trains two models; genes the models flag are switched off to test the prediction. Bottom lane - clinic: DNA from a patient's archived tumour sample goes to Model 2, which fills in the RNA and protein picture. Model 1 then identifies the active resistance programme, and the report informs the oncologist's choice of chemotherapy.
Who does what, where
| The Australian National University (ANU) | Grows and exposes the cell lines to chemotherapy, measures proteins, tests predicted genes by switching them off and builds the AI models. |
| National Computational Infrastructure (ANU) | AI models will be trained on the National Computational Infrastructure. |
| JCSMR proteomics node (ANU) | Proteins will be measured through mass spectrometry on the latest instruments at the JCSMR proteomics node |
| University of Kent (UK) | Supplies resistant cancer cell lines and their parental lines from its Resistant Cancer Cell Line collection. |
| PacBio | Long-read sequencing of whole genomes and full-length RNA, with study-design expertise. |
| Canberra Hospital and Victorian Cancer Biobank | Archived patient tumour samples with matched normal tissue and clinical outcome data, used to test the models on patients' tumours. |
Figure 2 | Who does what, where. Boxes are the partners and what each does; arrows show what moves between them. Cell lines come from the University of Kent, long-read sequence data from PacBio and archived patient tumour samples from the Victorian Cancer Biobank to the John Curtin School of Medical Research, where the atlas and models are built, with computing on the National Computational Infrastructure, and then released openly.
What it will deliver
An open-access atlas, with openly licensed AI models and software. Models small enough to run on hospital hardware. Because the second model needs only tumour DNA, the approach is intended to extend to archived tumour collections. The tool will help to support oncologists' decisions, not aim to replace them.
At a glance
| Where | John Curtin School of Medical Research, Australian National University |
| How long | Three years |
| Cell lines | Resistant and sensitive to chemotherapy |
| Measurements | Four per line: whole genome, full-length RNA, single-cell gene activity, proteins |
| AI models | Two: one reads resistance programmes, one works from tumour DNA alone |
| Testing | Gene knockout in the laboratory; patient tumour samples |
| Outputs | Open atlas, AI models and software, free to use |
Contact: Eli Niktab, John Curtin School of Medical Research, Australian National University. eli.niktab@anu.edu.au