Extended Data Fig. 5: Comparison and interpretation of the neural-network-based instance segmentation frameworks. | Nature Biotechnology

Extended Data Fig. 5: Comparison and interpretation of the neural-network-based instance segmentation frameworks.

From: De novo and somatic structural variant discovery with SVision-pro

Extended Data Fig. 5: Comparison and interpretation of the neural-network-based instance segmentation frameworks.The alternative text for this image may have been generated using AI.

a, Comparison of the accuracy (y-axis) on validation dataset among the five models (x-axis). The models are arranged based on their parameter sizes. b, the network architecture of the default Lite-Unet model. c, A heatmap to illustrate the Feature Ablation interpretation of the Lite-Unet model. Positives values (in green) indicates positive attrition to the specific prediction while negative values are shown in red. d, Using Grad-Cam to generate attribution maps of each layer in Lite-Unet.

Back to article page