Abstract IA010: From Multi-Omics to Autonomous AI Agents: AI-HOPE-Pancreas for Precision Oncology
Enrique Velazquez VillarrealAbstract
Background:
Pancreatic ductal adenocarcinoma (PDAC) is characterized by extensive molecular heterogeneity, complex pathway interactions, stromal remodeling, and variable responses to systemic therapy. Although KRAS and TP53 represent major genomic drivers, the broader signaling architecture underlying age- and treatment-specific differences remains incompletely understood. We developed a conversational artificial intelligence (AI) agent framework to enable rapid, clinically contextualized interrogation of multiple oncogenic pathways and establish a foundation for autonomous AI-driven multi-omic precision oncology in pancreatic cancer.
Methods:
Clinical and genomic data from 184 PDAC tumors were analyzed after stratification by age at diagnosis and gemcitabine exposure. A specialized AI-HOPE-Pancreas conversational artificial intelligence agent, developed at the Velazquez-Villarreal Lab and targeting the RTK-RAS, MAPK, TP53, PI3K, TGFβ, and JAK/STAT pathways, was used to dynamically construct clinically defined cohorts and perform pathway- and gene-level analyses. AI-generated findings were subsequently confirmed using conventional statistical approaches. Molecular alteration patterns and their associations with overall survival were evaluated across age- and treatment-specific subgroups.
Results:
Distinct pathway architectures emerged across clinical contexts. Within RTK-RAS/MAPK signaling, ERBB2 & RET mutations were enriched among gemcitabine-treated late-onset tumors, whereas early-onset disease demonstrated treatment-associated differences involving CACNA2D family genes, FLNB, and TP53. TP53 alterations were more frequent in gemcitabine-treated early-onset compared with late-onset PDAC, while PI3K pathway alterations were enriched in late-onset gemcitabine-treated tumors. TGFβ alterations occurred in approximately one-quarter to one-third of tumors across subgroups and were predominantly driven by SMAD4. TGFBR2 mutations were significantly enriched in gemcitabine-treated versus untreated late-onset PDAC. In contrast, genomic alterations affecting JAK/STAT signaling were uncommon and showed no significant age- or treatment-associated differences. Importantly, among late-onset patients who did not receive gemcitabine, the absence of RTK-RAS/MAPK, TP53, or PI3K pathway alterations was associated with improved overall survival.
Conclusions:
Integrated interrogation of multiple signaling pathways reveals that PDAC molecular architecture is strongly dependent on clinical context, including age at diagnosis and treatment exposure. These findings extend molecular characterization beyond individual driver genes and demonstrate how specialized conversational AI agents can rapidly integrate data with potential prognostic and therapeutic relevance. This framework provides a foundation for expanding from genomic analyses toward multi-omic integration and increasingly autonomous AI agents capable of coordinating genomic, transcriptomic, microbiome, spatial & clinical data. These approaches may accelerate AI-guided precision medicine in pancreatic cancer.
Citation Format:
Enrique Velazquez Villarreal. From Multi-Omics to Autonomous AI Agents: AI-HOPE-Pancreas for Precision Oncology [abstract]. In: Proceedings of the AACR Conference on Pancreatic Cancer: New Frontiers in Biology and Therapeutic Development; 2026 Sep 25-28; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(18_Suppl_2):Abstract nr IA010.