Citations

Abstract Anthracycline-induced cardiotoxicity (AIC) remains a dose-limiting complication of cancer treatment, often leading to irreversible heart failure, yet effective cardioprotective therapies remain limited. Graph-driven drug repurposing offers a rapid pathway to identify cardioprotective therapies, but different computational models often generate divergent candidate lists that exceed experimental validation capacity. We applied a novel "Union-then-Filter" framework to prioritize therapeutics for AIC using AI-based approaches and evaluated candidates in zebrafish models. Three state-of-the-art graph learning models (TxGNN, CompGCN, and RLR) were trained on the PrimeKG to generate an initial candidate pool. A Large Language Model (LLM) agent automated evidence synthesis across 4,286 PubMed abstracts for the top 50 drugs from each model. Candidates were stratified using dual decision criteria and filtered against existing clinical trials. The pipeline narrowed 150 initial candidates to four novel candidates: Cysteine, Dasatinib, Tranilast, and Tretinoin. Prospective in vivo validation was subsequently conducted using an adult AIC (aAIC) zebrafish model. Phenotypic screening via echocardiography revealed that Cysteine, Tranilast, and Tretinoin significantly restored ejection fraction (EF) (overall one-way ANOVA: \(\:F\left(\text{4,95}\right)=47.66,\:\:p<0.0001\); Tukey's post hoc: \(\:\:\:p<0.01\) for all three treatments vs. DOX+ control) and normalized cardiac remodeling markers in doxorubicin-treated subjects (Kruskal-Wallis: \(\:{\chi\:}^{2}\left(4\right)=17.21,\:\:p=0.0018\); Dunn’s post hoc: \(\:\:\:p<0.01\) for all three comparisons vs. DOX control). This study demonstrates the translational utility of an LLM-integrated AI pipeline for drug repurposing and highlights zebrafish AIC as an efficient bridge between in silico prediction and in vivo phenotypic validation in cardio-oncology.

Abstract Background: Pancreatic ductal adenocarcinoma (PDAC) is the most lethal common cancer (5-year survival ∼13%), emerging through years-to-decades progression involving chronic fibro-inflammatory injury and precursor transformation processes including ADM and PanIN. KRAS-driven acinar-to-ductal metaplasia (ADM) is among the earliest experimentally tractable and potentially interceptable events; KRAS (mutated in &gt;90% of PDAC) has remained difficult to therapeutically target, particularly for prevention; emerging KRAS inhibitors are developed for treatment of established malignancy rather than early interception. We aimed to prioritize KRAS-axis interception candidates via AI-guided network analysis of TCGA-PAAD restricted to a curated KRAS-driven ADM/PanIN-initiation mechanism module. Methods: The 38-gene module (n=183 TCGA-PAAD samples; 37 genes available, PRSS2 absent) spans KRAS-MAPK, RTKs, TGF-β/SMAD, ECM, inflammation, transcription, tumor suppressors, and ductal/acinar genes from canonical pancreatic-cancer-progression biology, linking established-tumor transcriptomes to early-interception biology. A stability-selected Spearman-correlation network served as the network backbone (PCMCI+/NOTEARS extensions ongoing), paired with patient-level bootstrap filtering (B=200 resamples; ≥70% retention). A composite Interception Score I(v) = α·CB + β·Σ|w| + γ·U + δ·R integrating betweenness centrality, summed edge-weight influence, druggability (literature-curated drug-target prior), and bootstrap robustness ranked candidate interception nodes (unoptimized weights; tuning ongoing). Results: Spearman correlation yielded a 37-node, 248-edge network baseline; bootstrap resampling (B=200) evaluated 856 distinct candidate edges, of which 194 (22.7%) achieved ≥70% retention, defining the stable subgraph used for scoring. The Interception Score placed 7 KRAS-axis genes among the top-10 candidates: ERBB2 (0.953), KRAS (0.743), ERBB3 (0.742), MAPK1 (0.708), EGFR (0.675), MAP2K2 (0.605), and BRAF (0.589). Inflammation (IL6, 0.755), TGF-β/SMAD2 (0.726), and ECM/MMP2 (0.638) also ranked highly, recapitulating PDAC initiation mechanisms (stromal-epithelial crosstalk; inflammatory-fibrotic priming). ERBB2 (HER2), a clinically druggable RTK with approved inhibitors in other malignancies, emerged as the top-ranked candidate, supporting evaluation of HER2-axis signaling for interception in molecularly selected PDAC subsets. These findings align with canonical PDAC biology, supporting biological plausibility prior to formal weight optimization and external validation. Conclusions: AI-guided network analysis of TCGA-PAAD prioritizes biologically coherent KRAS-axis and stromal-inflammatory candidates with clinically available HER2-, EGFR-, MEK-, BRAF-, and KRAS-pathway inhibitors, supporting hypothesis generation for future pancreatic cancer interception studies. Stage-resolved analysis, causal-inference extensions, and sex-/age-stratified validation are ongoing. Citation Format: Jianfu Li, Yingyun Yang, Michael Wallace, Cui Tao, Yan Bi. AI-Driven Network-Based Discovery on TCGA-PAAD Identifies KRAS-Axis Drug-Repurposing Targets for Pancreatic Cancer Interception [abstract]. In: Proceedings of AACR Drug Discovery and Development (AACR D3) Conference; 2026 Jul 21-24; Boston, MA. Philadelphia (PA): AACR; Clin Cancer Res 2026;32(14_Suppl):Abstract nr A029.