Few-Shot Learning for Multi-Omics Disease Classification with a MAML-Based Mode
1
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
Few-shot learning offers a promising approach for disease classification in settings where labeled data are scarce. While widely exploredin cancer research, its application to non-cancer diseases and multi-omics data remains limited. In this study, we propose a MAML-basedfew-shot learning model, pre-trained on TCGA data from four distinct tissue types. We then evaluate its adaptability across three diseasecategories, including COVID-19, Cirrhosis, and HBV-HCC. Our results demonstrate that MAML consistently outperforms a baseline MLP,achieving higher PR-AUC and ROC-AUC for COVID-19 and Cirrhosis. However, for HBV-HCC, where disease characteristics closely alignwith the pre-training data, the baseline MLP exhibits slightly superior performance. These experimental results suggest that diseaseclassification is feasible even with multi-omics data collected under limited sampling conditions. Furthermore, this study identifies diseasegroups where the advantages of MAML can be effectively leveraged.
Abstract
Few-shot learning offers a promising approach for disease classification in settings where labeled data are scarce. While widely exploredin cancer research, its application to non-cancer diseases and multi-omics data remains limited. In this study, we propose a MAML-basedfew-shot learning model, pre-trained on TCGA data from four distinct tissue types. We then evaluate its adaptability across three diseasecategories, including COVID-19, Cirrhosis, and HBV-HCC. Our results demonstrate that MAML consistently outperforms a baseline MLP,achieving higher PR-AUC and ROC-AUC for COVID-19 and Cirrhosis. However, for HBV-HCC, where disease characteristics closely alignwith the pre-training data, the baseline MLP exhibits slightly superior performance. These experimental results suggest that diseaseclassification is feasible even with multi-omics data collected under limited sampling conditions. Furthermore, this study identifies diseasegroups where the advantages of MAML can be effectively leveraged.