Abstract T-lineage acute lymphoblastic leukemia (T-ALL) is an aggressive pediatric malignancy that hasn’t been fully characterized, partly due to the high prevalence of noncoding genomic alterations driving oncogenic deregulation. Identifying T-ALL subtypes is essential for downstream risk stratification and therapeutic strategy selection. Conventional methods for T-ALL characterization, such as immunophenotyping, cytogenetic analysis, fluorescence in situ hybridization (FISH), and targeted molecular assays, are often labor-intensive, time-consuming, and costly. Furthermore, many T-ALL cases harbor alterations in noncoding genomic regions, which are difficult to detect by standard diagnostic workflows. To address these challenges, we present OTTER (Optimal Transport-based Transcriptomics and gEnomics Representation), a novel multi-modal learning framework that integrates transcriptomics data and genomics data for accurate and cost-effective T-ALL subtyping. OTTER first processed single nucleotide variations (SNVs) and gene expression profiles, respectively, through attention modules to capture the most informative features. Then, it leveraged an optimal transport (OT) method to align and fuse heterogeneous omics modalities, capturing complementary biological information across data types. Based on this, the OT-derived cost matrix learned the cross-modal interdependencies by quantifying the cost of aligning one modality to the other. In this way, OTTER generated a shared latent representation that integrated both omics data, enabling a more comprehensive and coherent representation of each patient sample. Experimental results based on 1,300 T-ALL patients demonstrated that multi-omics integration via OTTER significantly outperformed single omics approaches in subtype classification across performance matrices. In addition, OTTER achieved a substantially higher performance than the baseline models (e.g., ensemble learning models and attention-based models). Furthermore, the embeddings derived from OTTER could more clearly separate different T-ALL subtypes in tSNE visualization compared to approaches without OT, highlighting its ability to uncover subtype-specific molecular features for T-ALL. In summary, OTTER is an accurate and cost-effective framework for T-ALL subtyping that could leverage multi-omics data for improved T-ALL characterization performance. Based on this, we believe OTTER will significantly improve downstream T-ALL patient risk assessment and personalized treatment design. Citation Format: Lusheng Li, Jieqiong Wang, Shibiao Wan, . OTTER: Optimal transport-based transcriptomics and genomics representation fusion for T-ALL subtyping 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 6904.
Li et al. (Fri,) studied this question.