Researchers from the University of East Anglia (UEA) have developed an advanced model that identifies at least three distinct subtypes of osteosarcoma, a rare and aggressive form of bone cancer, which could have a significant impact on clinical trials and patient treatment. Funded by Children with Cancer UK, the breakthrough uses a machine learning technique called Latent Process Decomposition (LPD) to analyze genetic data and categorize patients more accurately. The research is published in Briefings in Bioinformatics.
Osteosarcoma, which primarily affects children and teenagers, has long been treated with a single approach, using chemotherapy and surgery. But only some patients respond well to these treatments.
“Since the 1970s osteosarcoma has been treated using untargeted chemotherapy and surgery, which sometimes results in limb amputation as well as the severe and lifelong side effects of the chemotherapy,” said lead author Darrell Green, PhD, a lecturer and group leader at UEA Norwich Medical School.
Historically, clinical trials for osteosarcoma have been difficult to conduct due to the heterogeneity found in osteosarcoma, and many have been deemed failures, as the total number of responders to new therapies remained low. However, this new research found that previous trials may not have been assessed accurately given what the team uncovered.
“The new medicines were not a total ‘failure’ as was concluded; rather, the drugs were not successful for every patient with osteosarcoma but could have become a new treatment for select patient groups,” Green noted.
By grouping patients based on their genetic profiles using LPD, doctors could more effectively match treatments to individual subtypes of the disease, potentially improving the chances of success in clinical trials. The model detects “functional states” within tumors, recognizing the differences that exist even within a single tumor, which previous models didn’t do.
“We have used a more sophisticated unsupervised Bayesian method, which considers individual tumor sample heterogeneity (where previous methods do not),” the researchers wrote.
Using LPD, the investigators defined three distinct subtypes of osteosarcoma, one of which showed a particularly poor response to the standard chemotherapy regimen, MAP (methotrexate, doxorubicin, and cisplatin). The researchers hope that this new method will allow for more targeted treatments, ultimately improving patient outcomes.
“When patients can be treated using targeted drugs specific to their cancer subtype, this will facilitate a move away from standard chemotherapy,” Green added.
The research also identified a core set of eight genes consistently dysregulated in osteosarcoma, which could serve as potential biomarkers for future diagnostic and therapeutic purposes. These genes, some of which have previously been linked to poor prognosis, could help identify high-risk patients and guide treatment decisions.
Despite the promising results, the study’s researchers acknowledge the limitations of their work, particularly the small dataset used to develop the model and the challenges posed by limited biopsy material. However, they are optimistic that as more data becomes available, the LPD model will continue to improve.
The identification of osteosarcoma subtypes could change the landscape of clinical trials. With better patient stratification methods, new drugs could be tested with a higher likelihood of success. The hope is that this approach will lead to better survival rates, which have remained stagnant at around 50% for the past 45 years.
“In the future, clinical trials should group patients with osteosarcoma according to their subtype (based on gene expression) so that treatment is tailored to their disease,” the researchers noted. “We expect that this new diagnostic labeling followed by stratified treatment will significantly improve osteosarcoma survival.”
