Hirst David P, Térézol Morgane, Cantini Laura, Villoutreix Paul, Vignes Matthieu, Baudot Anaïs
Aix Marseille Univ, INSERM, MMG, Centuri, Marseille, France.
Institut Pasteur, Université Paris Cité, CNRS UMR 3738, Machine Learning for Integrative Genomics Group, Paris, F-75015, France.
Genome Biol. 2025 Jul 25;26(1):224. doi: 10.1186/s13059-025-03675-7.
Joint matrix factorization is popular for extracting lower dimensional representations of multi-omics data but loses effectiveness with limited samples. Addressing this limitation, we introduce MOTL (Multi-Omics Transfer Learning), a framework that enhances MOFA (Multi-Omics Factor Analysis) by inferring latent factors for small multi-omics target datasets with respect to those inferred from a large heterogeneous learning dataset. We evaluate MOTL by designing simulated and real data protocols and demonstrate that MOTL improves the factorization of limited-sample multi-omics datasets when compared to factorization without transfer learning. When applied to actual glioblastoma samples, MOTL enhances delineation of cancer status and subtype.
联合矩阵分解在提取多组学数据的低维表示方面很受欢迎，但在样本有限时会失去有效性。为了解决这一局限性，我们引入了MOTL（多组学迁移学习），这是一个通过针对从大型异质学习数据集中推断出的潜在因子，为小型多组学目标数据集推断潜在因子来增强MOFA（多组学因子分析）的框架。我们通过设计模拟和真实数据协议来评估MOTL，并证明与无迁移学习的分解相比，MOTL改善了有限样本多组学数据集的分解。当应用于实际的胶质母细胞瘤样本时，MOTL增强了癌症状态和亚型的划分。
