In an advance for AI in drug discovery, SOPHiA GENETICS, in collaboration with AstraZeneca, recently presented data generated using multimodal machine learning models to identify subgroups of stage IV non-small cell lung cancer (NSCLC) patients who could most benefit from a specific treatment—the addition of tremelimumab to durvalumab and chemotherapy.
The team conducted a retrospective, multimodal analysis of the POSEIDON Phase III clinical trial (NCT03164616). This trial originally demonstrated that the combination of tremelimumab, durvalumab, and chemotherapy significantly increases progression-free survival (PFS) and overall survival (OS) versus chemotherapy in patients with metastatic NSCLC, which lead to approval of this regiment globally in NSCLC.
The SOPHiA GENETICS study used multimodal machine learning models to analyze clinical, biological, genomic, and imaging data, pinpointing patient subgroups who are most likely to benefit from the combination treatment.
“The objective of this project was to apply a multimodal machine learning approach to the results of this clinical trial to identify specific patient subpopulations that may get a higher predicted benefit from the addition of tremelimumab to durvalumab and chemotherapy,” Philippe Menu, MD, PhD, told Inside Precision Medicine. He is CMO and CPO at SOPHiA.
He added that, “The results of this re-analysis were positive, in the sense that we could identify a group of patients across lung cancer histologies that are predicted to show a 25% relative risk reduction of death with the addition of tremelimumab to the durvalumab and chemotherapy backbone.”
The research highlighted signatures identifying patients with non-squamous metastatic NSCLC who may derive higher OS benefit from the addition of tremelimumab to durvalumab plus chemotherapy in the first-line treatment setting.
In particular, EGFR wild-type, FGFR3 wild-type, CDKN2A wild-type, KRAS mutation, and STK11 mutation, as a signature, were identified as being associated with a higher OS benefit. These findings, the researchers say, could provide an exploration avenue towards a more tailored approach to patient care.
Menu explained, “The full data stack of the clinical trial was augmented through the generation of radiomics data from the medical images. Briefly, CT scans were digitally re-analyzed to focus only on the tumors in the images, and about 200 different data metrics known as ‘radiomics features’ were generated. In that sense, the images were translated into data.
Then a multimodal machine learning model was applied across the combined data stacks of genomics, clinical, laboratory, radiology and radiomics data to identify subpopulations of patients that are predicted to draw a higher overall survival benefit from the addition of tremelimumab to the durvalumab-chemotherapy backbone.”
“Our collaboration with AstraZeneca represents a major step forward in personalized oncology. Non-small cell lung cancer remains one of the most challenging cancers to treat due to its complex biology and the late stage at which it is often diagnosed,” said Jurgi Camblong, PhD, co-founder and CEO of SOPHiA GENETICS.
“This study harnesses the power of multimodal data and advanced AI to identify which patients are most likely to benefit from specific therapies. By tailoring treatment strategies based on a patient’s unique multimodal profile, we aim to improve outcomes and offer new hope to those battling this difficult disease.”
The study was presented as a poster by Ferdinandos Skoulidis, the department of thoracic medical oncology, University of Texas MD Anderson Cancer Center at ESMO 2024, in Barcelona last week.
