‘Molecular twin’ platform could expand use of precision medicine in cancer care
Click Here to Manage Email Alerts
Researchers used an artificial intelligence and precision medicine approach that identifies biomarkers that appear superior to standard tests for predicting pancreatic cancer survival, findings in Nature Cancer showed.
Clinical data — such as lab tests and pathology findings — are combined with multi-omic molecular data (eg, DNA and RNA) to create a so-called “molecular twin,” a virtual replica of the patient and his or her tumor.
The information gleaned from this effort will be used to construct a database of “scientific stunt doubles” that can help identify optimal therapies and potentially provide insights into how cancer will affect a specific patient, according to Dan Theodorescu, MD, PhD, director of Cedars-Sinai Cancer and corresponding author on the study.
The initiative — which investigators hope one day could improve treatment for all patients with cancer — also could facilitate research among underserved populations and allow for evaluation of how ethnic and racial disparities impact tumor biology and treatment.
“We’ve made tremendous progress in cancer,” Theodorescu told Healio. “We’ve turned a corner by really gaining a detailed understanding of the molecular mechanisms using a variety of technologies. Sometimes, you work and work, and then suddenly you reach an inflection point. I believe that’s where we are. The next few years will see transformational progress facilitated by the adoption of artificial intelligence in biological/medical science.”
Cancer and the host
The paper published in Nature Cancer outlined how Theodorescu and colleagues developed and tested a platform that used clinical and molecular analytes from the tumor in this proof-of-principle study. The team used cases of pancreatic ductal adenocarcinoma.
Researchers performed the analysis using plasma and tumor tissue samples from 74 patients with stage I or stage II resectable pancreatic ductal adenocarcinoma.
They conducted targeted, next-generation sequencing of DNA and whole-transcriptome RNA sequencing, tissue proteomics, plasma proteomics and computational pathology to generate individual omics analytes.
They then applied AI and machine-learning approaches to these data to determine which set of multi-omic biomarker panels predicted survival.
A multi-omic model that integrated all 6,363 feature datapoints across all molecular and clinical analytes provided the best prediction of survival among patients with surgically resected pancreatic ductal adenocarcinoma. Notably, all single-omic and multi-omic modeling used in the platform significantly outperformed CA-19-9 — the only FDA-approved pancreatic cancer test — as a prognostic tool.
Theodorescu said he has hypothesized for some time that an effective prediction tool would need to characterize and analyze not just tumor information, but the interplay between the cancer and the individual patient.
“When a patient has cancer, the body responds. You need to capture that response to be able to fully understand what is happening and formulate optimal patient management,” Theodorescu said. “For a long time, we only looked at the cancer. This multi-omics approach breaks that rule because it looks at the analytes of the host and the cancer and uses AI to examine the entire dataset. That’s a big difference — my hypothesis is that unless you study this data wholistically, you’re not going to have a good predictive marker.”
In many cases, patients with the same cancer at the same stage may have very different outcomes, Theodorescu said. This casts doubt on the reliability of predictions made solely based on cancer type or stage.
“Our paper takes patients with the same cancers at similar stages and proves that if you look at both the cancer and the host, you get better predictions,” Theodorescu said.
Identifying ‘optimal’ treatment
The proof-of-principle use of this platform in pancreatic ductal adenocarcinoma establishes a framework for developing highly informative, multi-omic biomarker panels that can be applied to scientific discovery, clinical practice and prognostication across cancer types, Theodorescu said.
Researchers envision using these “molecular twins” to create a database of de-identified clinical and molecular information from thousands of patients, allowing for extensive research into individualized risk assessments and treatment strategies, as well as insights into why some patients develop resistance to certain therapies.
Theodorescu envisions this platform delivering actionable information on many different cancer types.
“It’s very unlikely that this will only work in pancreatic cancer,” he said. “We chose pancreatic cancer for obvious reasons. It’s very lethal and we do major operations on patients, yet most people don’t benefit. If we determine with good confidence that a patient isn’t going to benefit, we need to offer them clinical trials rather than exposing them to a treatment with significant mortality and morbidity.”
In the future, the platform will include larger numbers of patients and will be applied prospectively, with several time points for longitudinal molecular profiling.
The platform provides flexibility and possibility for expansion around a “computational replica of a patient,” the researchers wrote in the Nature Cancer paper. Ultimately, it will integrate analytes from wearables, microbiomes and computational radiology.
Theodorescu discussed the “decision matrix” that the platform provides, noting that it has the potential to inform treatment decisions in terms of outcomes from surgery, chemotherapy or clinical trials.
“Since the platform can be trained to predict nearly all clinical outcomes and treatments, it will eventually be able to assign the patient to the optimal initial and follow-up treatments,” he said. “Eventually, molecular tests that encompass tumor and patient information will form the basis of core decision nodes in clinical practice guidelines.”
For more information:
Dan Theodorescu, MD, PhD, can be reached at Theodorescu Lab, 8700 Beverly Blvd., Davis Building, Room 3057, Los Angeles, CA 90048; email: dan.theodorescu@cshs.org.
Collapse
Click Here to Manage Email Alerts