Clinical utility analysis of a computational reasoning system on large-scale tumor genomic data.

Abstract

1625 Background: Comprehensive tumor molecular testing has become routine in clinical oncology. However, interpretation of complex molecular profiles in clinical practice can be challenging and subjective. The Digital Drug Assignment (DDA) system is a knowledge-graph-based computational method that automates reasoning at the patient level and scores molecularly targeted agents (MTAs) based on the full tumor genome. This approach was predictive of relative benefit of the agents as used in the SHIVA01 trial (NPJ Precis Oncol. 2021 5:59) and in real-world setting (NPJ Precis Oncol. 2025 9:159). Here, we assessed the potential clinical utility of DDA by analyzing large-scale tumor genomic data. Methods: Tumor molecular profiles of 20,923 patients of the MSK-CHORD real-world clinicogenomic dataset (Nature. 2024 636:728-36) were processed by DDA. DDA classified all alterations and scored all associated MTAs based on a functional and pharmacological evidence network. Previous research evidenced that high-score MTAs (1000 ≦ DDA score) predict increased clinical benefit. Utility of DDA was assessed by the proportion of patients with at least one high-score MTA option associated with their molecular profiles. Results: Tumor molecular complexity is evidenced by the number of driver alterations, with an average of 4.6 (range: 1 – 45) drivers per tumor (Table). 75% of all tumors were associated with high-score MTAs, 55% had on-label options. Availability of high-score options was high in all tumor types, with the exception of prostate tumors (24%). Only 3% of pancreatic cancer patients were associated with on-label high-score MTA options. Of the patients with known MTA treatments (n = 14,168), only 37% were administered a high-score MTA during their treatment course (Table). Conclusions: Our analysis recapitulates the presence of multiple driver alterations, highlighting the need for methods like DDA that consider the totality of molecular profiles for assigning MTAs, rather than relying on identifying single ‘actionable biomarkers’. The substantial gap between the proportion of patients who did receive and those whose tumor genomes were associated with high-score MTAs indicates a strong potential for improving outcomes in precision oncology. Moreover, the proportion of high-score off-label (i.e., approved in other oncological indications) MTAs highlights the potential of DDA in drug repurposing. Tumor site Mean no. of alterations (±SD) Mean no. of drivers (±SD) HS MTAs associated* HS MTA administered* HS developmental MTAs associated* HS off-label MTAs associated* HS on-label MTAs associated* All (20,923) 9.9 (11.8) 4.6 (3.9) 75 37 74 71 55 Breast (5,164) 8.1 (8.1) 4.3 (3.4) 90 57 90 89 86 Colorectal (4,565) 13.2 (18.0) 6.0 (5.5) 69 18 67 60 46 Lung (5,932) 11.6 (10.2) 4.6 (3.2) 85 57 83 84 79 Pancreas (2,619) 6.2 (6.0) 3.6 (2.3) 86 2.5 86 86 3 Prostate (2,643) 7.4 (9.5) 3.6 (3.3) 24 6.6 21 11 7 HS: high-score. *% of patients.