The field of multi-omics combines extensive data from genes, proteins, and biomolecules to build a comprehensive picture of biological processes and diseases. Modern methods of laboratory and clinical analyses can generate enormous amounts of data from a single patient, and sophisticated AI and machine-learning tools are being developed to examine these datasets. Cedars-Sinai Cancer Center has created a unique precision medicine and artificial intelligence (AI) tool called the Molecular Twin Precision Oncology Platform (MT-POP) to examine large datasets from cancer patients and identify which specific biomarkers can accurately predict cancer survival.
Conceptually, the MT-POP differs from other precision medicine and AI approaches since it is based on the fundamental hypothesis that a comprehensive evaluation of both the patient and the tumor is the key to determining the diagnostic and therapeutic success in cancer medicine. This is because we now have ample scientific evidence that diagnostic and therapeutic success depends on understanding the delicate and bespoke interaction between the patient and their cancer and the ability to assess that dynamic relationship in real-time.
Credit. Author
How can the development of biomarker predictors of cancer therapy outcome be improved?Incorporating multi-omic data sources and processing these through artificial intelligence tools has shown that predicting pancreatic cancer outcome after surgery is superior to that provided by single-omic markers currently for such predictions.
Dan Theodorescu
Using AI to predict pancreatic cancer outcomes
In a study published in Nature Cancer, researchers examined large sets of data from surgically resected pancreatic cancer patients generated from both the patients and their tumors. This data included demographic information, clinical evaluation and tests, surgical pathology results, computational pathology results, and molecular analytes, such as DNA, RNA, proteins, and lipids.
Multiple independent machine learning models were developed, and researchers discovered a panel of 6,363 data points across all of the clinical and molecular information that proved the best predictor of survival in patients with surgically resected pancreatic cancer. This model accurately predicted disease survival in 87% of patients.
Democratizing precision medicine
A drawback to comprehensive approaches to collecting and analyzing large amounts of data from cancer patients is the cost of testing and resource requirements. This has limited such approaches to healthcare markets with the largest financial and social resources. Addressing this is a unique characteristic of the MT-POP architecture: its ability to use AI to identify a parsimonious composition of analytes that can optimize its predictive models. The Nature Cancer paper describes how researchers used AI to streamline their data, removing the least informative features while keeping good performance of the model. This resulted in a simplified model in which only 589 data points led to similarly robust predictions. Further examination of the data showed that analyzing proteins from a patient’s plasma was the most accurate predictor of disease survival.
The comprehensive and simplified plasma-protein models outperformed the only Food and Drug Administration-approved pancreatic cancer test, CA 19-9. These findings were validated using independent datasets from The Cancer Genome Atlas, Massachusetts General Hospital, and Johns Hopkins University.
This research can expand the availability of precision medicine to pancreatic cancer patients in locations that lack access to resources and advanced technology. Reducing the resource, cost, and time requirements for comprehensive testing can help to address global disparities in cancer care, offering the hope that precision medicine can be democratized.
Creating virtual replicas of patients
The MT-POP was launched at Cedars-Sinai in 2021. It was initially developed for pancreatic cancer with the vision of eventually expanding it across most cancers. The original aim was to improve the ability to predict which patients would benefit from the removal of their pancreas, a major procedure associated with significant morbidity.
A virtual model of a patient, their Molecular Twin, is created using AI to extract and process the multi-omic molecular data (DNA, RNA, protein) from the patient and their tumor tissues as well as multiple body fluids (sputum, saliva, urine, stool/microbiome). After consent for research on their samples is obtained, this comprehensive collection of blood and tissue samples is acquired from each patient seen in the clinic.
The MT-POP also analyzes patient’s clinical data, such as their medical history, physical and lab results, and data from AI-based computational analysis of their pathology and radiology tests. Clinical patient outcomes are also collected and stored in their Molecular Twin data. Importantly, the patients’ Molecular Twins can be progressively annotated and expanded with each visit based on the analysis of additional samples to improve their virtual models. This allows the MT-POP to progressively “learn” and improve its ability to predict the patient’s treatment response and outcome.
A community of molecular twins
The MT-POP will also promote the vision that thousands of virtual twins with known outcomes will eventually exist. This will create a large reference database that can be used to match new twins to the best treatment prospectively. The database will contain “de-identified” clinical and molecular patient information stripped of personal details, and it will be made available to researchers at Cedars-Sinai Cancer Center and other institutions globally. This database can offer insights into why some patients resist certain cancer treatments and allow the development of new cancer therapies.
By its widespread use, the MT-POP will also begin to contain important patient characteristics such as race, sex, gender, and geography. Ultimately, it will have the intelligence to determine which factors lead to a specific patient outcome. It will also aid patients and their physicians by ensuring rapid determination of patient eligibility for novel clinical trials. Over 1,000 patients have been enrolled by Cedars-Sinai Cancer Center into the Molecular Twin. More patients are planned to be added, with a goal of 1,000-2,000 patients each year.
What the future holds
While pancreatic cancer was the original focus of this research, the Molecular Twin will be an ever-increasing resource to aid in discovering biomarkers for other common cancer types, leading to the development of new treatments. When asked, most patients at Cedars-Sinai Cancer Center agree to participate in the Molecular Twin program. In addition to updating samples and data, investigators are constantly adding other types of data, such as samples of the gut microbiome and data from wearable devices. Our study has shown the superiority of incorporating multi-omic data sources in predicting pancreatic cancer outcomes after surgery compared to relying solely on single-omic analytes such as CA 19-9, the standard of care marker currently used for such predictions.
The eventual goal is for every Cedars-Sinai Cancer Center patient to receive a comprehensive Molecular Twin report that will allow their physicians to predict best how the patient’s specific cancer might respond to different treatments. Physicians will receive customized treatment information, including standard therapy options or information to help select the best clinical trial for a patient. This is an exciting era of precision medicine, and we are hopeful that the Molecular Twin Precision Oncology Platform will fuel progress in this field by stimulating research, improving patient treatments, and addressing global health disparities.
🔬🧫🧪🔍🤓👩🔬🦠🔭📚
Journal reference
Osipov, A., Nikolic, O., Gertych, A., Parker, S., Hendifar, A., Singh, P., … & Theodorescu, D. (2024). The Molecular Twin artificial-intelligence platform integrates multi-omic data to predict outcomes for pancreatic adenocarcinoma patients. Nature cancer, 5(2), 299-314. https://doi.org/10.1038/s43018-023-00697-7
