Deep convolutional neural networks for annotating gene expression patterns in the mouse brain

BMC Bioinformatics

Table 1 Overall AUC values achieved using different features across all ontology levels

	Voxel	BOW	L6	L9	L12	L16	L18
E11.5	0.759	0.841	0.898	0.903	0.905	0.898	0.876
E13.5	0.768	0.800	0.880	0.886	0.889	0.881	0.859
E15.5	0.754	0.768	0.870	0.875	0.876	0.871	0.840
E18.5	0.764	0.873	0.893	0.901	0.905	0.898	0.873
Overall	0.761	0.820	0.885	0.891	0.894	0.888	0.862

“BOW” and “Voxel” denote the performance achieved by the bag-of-words and grid-voxel level data, respectively. “Lx” denotes the performance of deep convolutional features, where “x” indicates the network layer from which the features were extracted. Note that L12 achieved the highest overall AUC in comparison with those achieved by other layers. BOLD- L12 achieved the highest overall AUC in comparison with those achieved by other layers.

ISSN: 1471-2105