Abstract 7: Molecularly-informed prediction of treatment efficacy in the GENIE BPC NSCLC cohort using computational reasoning.

Abstract Background: Digital Drug Assignment (DDA) is a computational reasoning model that scores cancer therapies based on the complete molecular profile of a tumor, and stratifies them by predicted efficacy (Petak et al., 2021). In a prior study of 111 lung cancer patients, DDA-derived high-score molecularly targeted agents (MTAs) were associated with improved clinical outcomes (Dirner et al., 2025). Here, we extend this analysis to the GENIE BPC NSCLC cohort to assess the broader clinical validity of DDA. Methods: From the GENIE BPC NSCLC cohort data available on Synapse, we included 1,078 patients with a single-sample genomic profile, available primary treatment data and survival outcomes (total 2,103 treatment lines, therapies included: afatinib, erlotinib, osimertinib, crizotinib, nivolumab, pembrolizumab, atezolizumab, bevacizumab+chemo, ramucirumab+chemo; and chemotherapy alone). DDA scores were generated for all cases, and the individual score of the administered MTAs (incl. immune checkpoint inhibitors) was used to stratify outcomes into low (0), intermediate, and high DDA-score (≥1000) tiers. Progression-free survival (PFS, by imaging) and overall survival (OS) were analyzed using Kaplan-Meier statistics. Results: Median PFS and OS differed significantly across DDA score tiers, increasing with higher scores (see table). Intermediate-tier drugs had similar mPFS values as chemotherapies (3.9 vs 4.2 months). Six-month PFS and twelve-month OS rates increased with DDA-tiers and were all significantly different by χ2 test. DDA-high therapies provided greater benefit across treatment types than lower-score counterparts. Conclusions: Across a large, real-world NSCLC cohort, DDA effectively distinguished therapies with higher clinical efficacy based on the full molecular profile of each patient. These results reinforce the potential of DDA to enhance personalized treatment selection based on NGS diagnostics in precision oncology. Citation Format: Barbara Vodicska, Eniko Kispeter, Dora Lakatos, Gabor Gy Kalmar, Robert Doczi, Dora Gorog-Tihanyi, Anna Dirner, Reka Szalkai-Denes, William T. Beck, Arkadiusz Z. Dudek, Christophe Le Tourneau, Istvan Petak. Molecularly-informed prediction of treatment efficacy in the GENIE BPC NSCLC cohort using computational reasoning abstract. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 7.