Dominant Regulatory Axis Detection links baseline AML regulatory state to drug-specific functional response across independent cohorts
Year: 2026
Type: preprint
Abstract: This record reports a frozen functional-response evaluation of Dominant Regulatory Axis Detection (DRA) in acute myeloid leukemia. Historical FIMM analyses supplied a prespecified hypothesis record containing both supported and falsified drug-response predictions. An independently qualified MERGE cohort was then analyzed under a locked contract: DRA scoring was completed outcome-blind, two drug-response hypotheses were tested, and both satisfied their frozen directional and multiplicity-controlled criteria. Prespecified robustness analyses were conducted only after the primary verdict was frozen. BeatAML2 was subsequently used as an explicitly post-lock mechanistic-context cohort to examine whether published AML differentiation-state scores accounted for the observed functional associations. The principal conclusion is that baseline regulatory state can carry reproducible, drug-specific functional information, while predictive association and biological mechanism remain separable validation layers. The record retains contradictory and negative evidence, including the unresolved NFκB–venetoclax relationship. This version has not undergone journal peer review. Abstract Baseline transcriptional state may contain functional information about drug response, but retrospective association analyses can blur hypothesis generation, validation, and mechanistic interpretation. We evaluated whether Dominant Regulatory Axis Detection (DRA), a low-dimensional framework for relative regulatory-state analysis, could support prospectively frozen functional-response hypotheses in acute myeloid leukemia (AML). Historical FIMM analyses provided a prespecified hypothesis record that included both supported and falsified drug-response predictions. We then qualified an independent MERGE cohort without inspecting DRA-response associations, froze two NFkB-axis hypotheses, generated the DRA predictor outcome-blind, and tested ex vivo response in 24 newly diagnosed AML specimens. Higher baseline NFkB state predicted greater rapamycin/sirolimus sensitivity (beta=-59.336 AUC units per SD; q=0.00457) and greater sorafenib resistance (beta=54.524; q=0.00457); both hypotheses satisfied the locked criteria. Rank-based, leave-one-patient-out, HC3, and influence analyses preserved the directional conclusions. In a separate post-lock BeatAML2 analysis of 277 initial-diagnosis AML patients, NFkB state was embedded in a differentiation structure characterized most strongly by higher Monocyte-like, lower Progenitor-like, and higher cDC-like scores. Full six-state adjustment attenuated but retained the negative NFkB-rapamycin association, whereas the NFkB-sorafenib coefficient reversed after differentiation adjustment. These results indicate that baseline regulatory state can carry reproducible, drug-specific functional information while the biological context underlying that information differs by treatment. Predictive regulatory-state information and biological mechanism should therefore be treated as separable validation layers. Version 1.1Presentation and documentation revision. Added three figures visualizing the locked MERGE functional-response associations, the evidence architecture, and post-lock BeatAML2 coefficient context; clarified selected methodological and interpretive wording; corrected minor citation/presentation issues. The frozen MERGE hypotheses, primary analyses, coefficients, multiplicity procedure, and primary verdicts are unchanged. No post-lock analysis was used to redefine the original confirmatory outcomes.
Author: Vlahopoulos, Spiros
Language: English
Subfield: Hematology
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7224996573