What question did this study set out to answer?

This research aims to characterize FGFR3-TACC3 fusion tumors through multi-omics analysis and machine learning to identify potential therapeutic targets.

April 5, 2026

Abstract 976: FGFR3-TACC3 fusion: multi-omics and machine learning characterization across tumor types

Key Points

This research aims to characterize FGFR3-TACC3 fusion tumors through multi-omics analysis and machine learning to identify potential therapeutic targets.
Analyzed transcriptomic data from TCGA cohorts comparing FGFR3-TACC3-positive to wild-type tumors.
Identified differentially expressed genes and intersected with druggability databases.
Analyzed 58 FGFR3-TACC3 fusion samples against 9,642 non-fusion cases for mutation analysis.
Applied machine learning techniques including a neural network and random forest to classify fusion profiles.
FGFR3-TACC3 fusion tumors showed distinct upregulation of 1,984 and downregulation of 2,504 genes.
Pathway analysis revealed enrichment in RTK/MAPK signaling and mitotic spindle assembly.
Identified 48 druggable genes linked to oncogenic pathways.
Machine learning achieved an AUC of 0.85 with 79% accuracy in distinguishing fusion from non-fusion profiles.

Abstract

Abstract Background: The FGFR3-TACC3 (FGFR3-TACC3) fusion is a recurrent oncogenic alteration observed in gliomas, urothelial, and head and neck cancers. This fusion drives constitutive FGFR3 kinase activation and disrupts mitotic spindle organization through TACC3, leading to uncontrolled proliferation and transcriptional reprogramming. To elucidate the molecular landscape and therapeutic vulnerabilities of FGFR3-TACC3-positive tumors, we integrated genomic, transcriptomic, and computational analyses. Methods: Transcriptomic data from TCGA cohorts were analyzed to compare FGFR3-TACC3-positive and wild-type tumors. Differentially expressed genes (fold change 2 or -2) were intersected with druggability databases to identify actionable targets. In parallel, we analyzed 58 FGFR3-TACC3 fusion samples and 9,642 non-fusion cases to determine recurrently mutated genes and enriched Gene Ontology (GO) terms. A combined neural network and random forest classifier was trained on mutational and functional features to discriminate fusion from non-fusion profiles. Results: FGFR3-TACC3 fusion tumors exhibited a distinct transcriptional and mutational signature, including 1,984 upregulated and 2,504 downregulated genes, with enrichment in RTK/MAPK signaling, mitotic spindle assembly, and vesicle transport pathways. Integration with pharmacologic databases revealed 48 co-expressed druggable genes, encompassing oncogenic kinases (FGFR3, EGFR, CDK4, PDGFRA, NTRK3) and neurotransmission-associated receptors (OXTR, ADORA1, GRIA3/4). Machine-learning analysis identified 9 recurrently mutated genes and 69 informative GO terms, achieving an AUC of 0.85 and accuracy of 0.79, highlighting a robust functional signature of FGFR3-TACC3 fusion tumors. Conclusions: This integrated multi-omics and machine-learning study suggests a functional, transcriptional, and pharmacologic landscape of FGFR3-TACC3 fusion-driven cancers. The combined mutation-GO-transcriptome framework uncovers potential druggable co-dependencies and provides hypotheses for rational combination strategies and precision therapies for FGFR3-TACC3-positive tumors that warrant experimental validation. Citation Format: Manuel Pedregal, Esther Cabañas Morafraile, Balázs Győrffy, Ester Garcia, Miriam Dorta, Bernard Gaston Doger de Spéville, Emiliano Calvo, Alberto Ocaña, Victor Moreno Garcia. FGFR3-TACC3 fusion: multi-omics and machine learning characterization across tumor types 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 976.

Ask AI

Helpful

Bookmark