A composite scaling network of EfficientNet for improving spatial domain identification performance DOI Creative Commons

Yanan Zhao,

Chunshen Long, Wenjing Shang

et al.

Communications Biology, Journal Year: 2024, Volume and Issue: 7(1)

Published: Nov. 25, 2024

Spatial Transcriptomics leverages gene expression profiling while preserving spatial location and histological images. However, processing the vast noisy image data in transcriptomics (ST) for precise recognition of domains remains a challenge. In this study, we propose method EfNST recognizing domains, which employs an efficient composite scaling network EfficientNet to learn multi-scale features. Compared with other relevant algorithms on six sets from three sequencing platforms, exhibits higher accuracy discerning fine tissue structures, highlighting its strong scalability operational efficiency. Under limited computing resources, testing results multiple show that algorithm runs faster maintaining accuracy. The ablation studies model demonstrate significant effectiveness EfficientNet. Within annotated sets, showcases ability finely identify subregions within structure discover corresponding marker genes. unannotated successfully identifies minute regions complex tissues elucidated their patterns biological processes. summary, presents novel approach inferring cellular organization discrete spots implications exploration function.

Language: Английский

A composite scaling network of EfficientNet for improving spatial domain identification performance DOI Creative Commons

Yanan Zhao,

Chunshen Long, Wenjing Shang

et al.

Communications Biology, Journal Year: 2024, Volume and Issue: 7(1)

Published: Nov. 25, 2024

Spatial Transcriptomics leverages gene expression profiling while preserving spatial location and histological images. However, processing the vast noisy image data in transcriptomics (ST) for precise recognition of domains remains a challenge. In this study, we propose method EfNST recognizing domains, which employs an efficient composite scaling network EfficientNet to learn multi-scale features. Compared with other relevant algorithms on six sets from three sequencing platforms, exhibits higher accuracy discerning fine tissue structures, highlighting its strong scalability operational efficiency. Under limited computing resources, testing results multiple show that algorithm runs faster maintaining accuracy. The ablation studies model demonstrate significant effectiveness EfficientNet. Within annotated sets, showcases ability finely identify subregions within structure discover corresponding marker genes. unannotated successfully identifies minute regions complex tissues elucidated their patterns biological processes. summary, presents novel approach inferring cellular organization discrete spots implications exploration function.

Language: Английский

Citations

0