Zephyr AI, a high-growth healthcare technology company committed to developing AI-Enabled Composite Biomarkers to expand the horizon of precision medicine in cancer drug development, presented a poster at the Annual Meeting for the American Society of Clinical Oncology (ASCO). Entitled “Evaluation of a novel signature for PARP inhibitor sensitivity prediction using real-world data,” the presentation demonstrates the game-changing patient insights that AI can derive from complex and diverse data sets that can be used to predict a patient’s sensitivity to a class of medicines called poly ADP ribose polymerase (PARP) inhibitors.
PARP inhibitors are currently authorized for use in the treatment of cancers that exhibit homologous recombination repair deficiencies (HRD+) or mutations in the breast cancer gene (BRCA) 1 or 2. These biomarkers are predominantly present in patients diagnosed with breast, ovarian, prostate, and pancreatic cancer and can predispose patients to PARP inhibitor sensitivity. However, recent data, across this class of medicines have revealed that their benefit-risk profile in late-line ovarian cancer patients using these biomarkers is insufficient to support their continued prescription. As such, these drug labels have recently been withdrawn creating a critical unmet need in late-line ovarian cancer patients to access this class of agent with the appropriate biomarker in place.
The Zephyr AI team has developed a novel, next-generation biomarker that could be used to identify late-line patients whose tumors harbor latent sensitivities to the PARP inhibitor olaparib, distinct from HRD+ and BRCA mutations. The team resolved this novel composite biomarker by using its custom machine learning (ML) foundation model developed over the last several years. This novel AI-enabled composite biomarker was validated using a real world cohort of ovarian cancer patients that had received olaparib in diverse drug combinations, typically, as a late-line therapy.
“Zephyr’s ML technology has the ability to precipitate a stepchange in patient enrichment approaches for new drugs emerging in oncology, an area where historically many drugs have failed or we have needed to refine our knowledge slowly and empirically,” said Jeff Sherman Co-Founder, Interim CEO, and Chief Technology Officer at Zephyr.
Also Read: Tempus Announces the Clinical Launch of its MRD Testing Portfolio
Zephyr validated its drug response predictions for this cohort using clinico-genomics and patient outcomes data. This analysis revealed that for patients predicted to be sensitive based on the Zephyr model a statistically significant improvement in real-world progression free survival from 24 months to more than 60 months (rw-PFS hazard ratio of 2.04 with a p-value less than 0.001) and a rw-overall survival improvement from 23 months to more than 60 months (rw-OS hazard ratio of 3.37 with a p-value less than 0.005) was observed. This profile was in contrast to analysis using the conventional HRD+ biomarker in the same cohort, where there was no statistically significant improvement in either of the clinically relevant, real-world end points reported.
“It is exciting to be able to share one of the applications of our ML foundation model that the team at Zephyr has been building. We anticipate that the assembly of enormous and complex multi-model data sets into a neural network that forms the basis of our foundation model will enable a new, composite approach to biomarkers broadly. This model will reveal novel insights across many mechanisms of action as we increase the breadth of training data provided to these models,” said Rachael Brake, PhD, Zephyr AI’s Chief Scientific Officer. “We believe that by looking beyond HRD+ and ‘BRCAness,’ we can uncover novel and latent drivers of PARP inhibitor sensitivity. This coincides with many next-generation PARP inhibitors entering clinical development and we are excited to partner with these drug development teams to increase the reach and probability of success of this important class of medicine.”
Source: Businesswire
