We appreciate the thoughtful comments by Zhu and colleagues (Letter to the Editor) on the utility of multiparametric MRI for detecting early response and acquired resistance of pancreatic cancer to KRAS inhibitor therapy (1). As they noted, the intratumoral heterogeneity captured by MRI can inform spatially guided biopsy and treatment strategies. Furthermore, analysis of MRI-derived heterogeneity features helps predict the development of therapeutic resistance, as demonstrated in previous studies of brain tumors (2).Clinical resistance to sotorasib and adagrasib—the two FDA-approved KRAS inhibitors targeting the G12C mutation—is well documented (3). To address this challenge, numerous clinical trials are exploring combinational strategies built around RAS inhibitor therapy, even prior to full regulatory approval of some RAS inhibitors (4). These combinations span a broad range of agents, including standard-of-care chemotherapies, immune checkpoint inhibitors, and therapies targeting RAS-associated pathways (4).Given this landscape, two key challenges remain in developing effective combination therapy. First, how can we efficiently identify the most effective regimen from the vast number of possible combinations, given that empirical selection in conventional trials may fail to capture synergistic effects? Second, once a regimen is selected, how can we determine its suitability for a specific patient according to their unique clinical and biological characteristics?In this context, statistical models calibrated to population-level data can help evaluate candidate combinations and identify optimal regimens for clinical testing. Once a promising regimen is identified, the next challenge is assessing its appropriateness for individual patients. A treatment effective in a trial cohort may not yield the best outcome for an individual due to interpatient variations in tumor biology, microenvironment, and treatment history. This is where the digital twin concept becomes particularly valuable (5). Recent studies have demonstrated the feasibility of this approach. For example, patient-specific digital twins have been developed for breast cancer using MRI data integrated with biology-based mathematical modeling. These twins successfully reproduced outcomes from several historical trials and identified theoretically optimal schedules on a patient-specific basis from >100 options (6).In summary, imaging-based measures of heterogeneity help predict resistance, whereas imaging-integrated modeling approaches at both the population and individual levels offer a promising path toward more effective combination therapies to overcome resistance to RAS inhibitor therapy.R. Zhou reports grants from NIH outside the submitted work. Y. Fan reports grants from NIH during the conduct of the study. No disclosures were reported by the other authors.