Abstract CT227: Quantum mechanics-based multi-tensor AI/ML discovery of a whole-transcriptome predictor of response and survival in patients treated with dabrafenib and trametinib in the tumor-agnostic small-cohort NCI-MATCH subprotocol H with valida

Abstract Background: The value of artificial intelligence and machine learning (AI/ML) in predicting outcomes in small cohorts is unclear. We previously have developed quantum mechanics-based multi-tensor AI/ML algorithms to overcome the limitations of typical AI/ML in, e.g., “skinny” data [doi: 10.1073/pnas.0530258100], and discovered and validated whole-genome and -chromosome predictors of response to treatment and survival in cohorts of, e.g., brain, lung, ovarian, and uterine cancer patients [doi: 10.1145/3624062.3624078]. The current study evaluated the ability of our method to derive clinically meaningful models from the transcriptomes of the tumor-agnostic small-cohort NCI-MATCH arm H [doi: 10.1200/JCO.20.00762], and whether such models could predict outcomes in cohorts from the Cancer Genome Atlas (TCGA). Methods: The multi-center subprotocol EAY131-H, which opened to enrollment in 8/2015 and completed accrual in 2/2018, evaluated patients with BRAF V600E/K/R/D -mutated tumors. The current study retrospectively included the 19 patients from the arm H that had tumor total RNA available. Models were derived from the transcriptomes alone, then tested as predictors by incorporating the patient labels, and associated with cellular mechanisms by using the transcript labels. Results: A 14,607/60,660-transcript predictor was discovered that is correlated with the arm H objective response rate (ORR) (Mann-Whitney-Wilcoxon P-value=2.2×10-2), as well as with progression-free survival (PFS) and overall survival (OS), with the Kaplan-Meier median PFS and OS differences of 8 and 20 months between the two groups of predictor-stratified patients (log-rank P-values=1.6×10-2 and 4.3×10-2), with both Cox PFS and OS hazard ratios =3.5. When evaluated in three separate cohorts in TCGA, the model was a similarly significant predictor of OS in melanoma, lung, and colorectal cancer, regardless of BRAF status. Response to dabrafenib and trametinib was associated with overexpression of transcripts involved in immune response and cell adhesion, including the major histocompatibility complex (MHC) genes and the neuroligins NLGN1-3,4X/Y. Resistance was correlated with cell cycle processes, such as overexpression of the c-Myc responsive CDCA7, the Cdc45-MCM-GINS complex genes, and the druggable TTK (MPS1). Conclusions: Our multi-tensor AI/ML algorithms derived a whole-transcriptome predictor of response and survival in the NCI-MATCH arm H. The model was able to predict OS in the three independent cohorts of melanoma, lung, and colorectal cancer in TCGA. This suggests that quantum mechanics-based algorithms can be used to derive clinically relevant predictors from small cancer cohorts. Citation Format: Orly Alter, Jong-In Park, Ping Guan, Asaf Zviran, Lyndsay N. Harris. Quantum mechanics-based multi-tensor AI/ML discovery of a whole-transcriptome predictor of response and survival in patients treated with dabrafenib and trametinib in the tumor-agnostic small-cohort NCI-MATCH subprotocol H with validation in three tumor types in TCGA [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 2 (Late-Breaking, Clinical Trial, and Invited Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_2):Abstract nr CT227.