GGE: General-purpose deep meta-learning for classification of human transcriptomes with limited data

Transcriptomic classification often struggles due to limited sample sizes compared to the high dimensionality of gene expression data. Researchers have developed General Gene Expression (GGE), a deep meta-learning framework intended to enhance robust classification in scenarios with scarce labeled data. By training on 5,220 biomedical prediction tasks derived from 1,779 human datasets, GGE develops an initialization model that captures shared biological patterns across diverse classification challenges.

This learned initialization offers two significant benefits. Initially, it allows for enhanced performance on new datasets even when provided with a limited number of labeled samples. Secondly, due to its training across varied prediction objectives, GGE can be applied to a wide array of biomedical problems. The results demonstrate that GGE surpasses conventional classifiers in data-scarce settings across numerous applications, including datasets generated through different RNA-seq platforms and preprocessing methods.

Furthermore, through attention-based analysis, the study identifies recurrent genes that contribute to GGE's performance across multiple biological objectives, offering insights into the shared factors influencing human biological states. In summary, GGE establishes itself as a versatile framework for human transcriptome-based classification in biomedical contexts where labeled data are limited.