A team of Stanford cancer scientists has been awarded funding from the Advanced Analysis for Precision Cancer Therapy (ADAPT) program to develop artificial intelligence (AI) models that integrate different data types and identify biomarkers to predict whether a patient will respond to a specific cancer treatment.
The models will incorporate longitudinal multimodal data, which is data derived from different sources when following the patient over a long period of time, to illustrate the complex relationships between tumor features, treatment response, and survival times. The multimodal data will include pathological, radiological, clinical, and molecular information.
Olivier Gevaert, PhD, Stanford Cancer Institute member and co-lead of the project, says, “We saw ADAPT as a big opportunity to develop new machine-learning models for multimodal data. The goal for our team specifically is to develop computational machine-learning AI tools that will extract multimodal biomarkers to help steer patients to novel treatments.”
The ADAPT program, housed in the Advanced Research Projects Agency for Health (ARPA-H) within the U.S. Department of Health and Human Services, selected 10 teams from academia and industry for a large-scale collaborative project that will determine effective therapies and strategies relative to a patient’s treatment course. Each team’s unique project will be a puzzle piece in ADAPT’s eventual findings.
The program will initially enroll lung, breast, and colon metastatic cancer patients. These patients will first receive the standard-of-care treatment and will have their tumors and blood samples tested throughout the trial. Unlike other clinical trials that assess data after the trial has concluded, ADAPT will continually evaluate patient data and adjust treatment in response.
Andrew Gentles, PhD, Stanford Cancer Institute member and co-lead of the project, says he has never seen this type of adaptive trial implemented before. He’s excited about the different types of data that will be available compared to a normal clinical trial.
He explains, “If somebody stops responding to a treatment, can we predict what’s happening in the evolution of that tumor and whether they should be moved to another specific treatment? The idea is to potentially change what treatment the patient receives as they go through the trial, and as we see how their tumor changes in various ways. In a standard clinical trial, patients are assigned to specific treatment arms, and the results are assessed at the end. In this case, the treatment will be adapting continuously throughout the trials, and the computational analysis is an intrinsic part of the trial.”
The Stanford team will start developing computational tools based on publicly available data and will switch to ADAPT’s trial data once it becomes available. The first patients will be enrolled within the next 12 months, and the project is expected to take six years to complete.
Gentles and Gevaert previously published a paper in Nature about using multimodal data for cancer biomarker discovery using AI models.
Gentles says, “This is a futuristic view of what we can do if we have many different types of data on the same patients and how that can be translated into the clinic because it’s typically not easy to get these multiple, different types of data on the same patient.”