A bag-of-words approach for Drosophila gene expression pattern annotation

BMC Bioinformatics

Table 2 Annotation performance in terms of AUC, macro F1, micro F1, sensitivity, and specificity for image groups in stage range 4–6.

Measure	# of terms	ML_LS	SVM	PMK_SIFT	PMK_comp
AUC	10	80.85 ± 0.74	78.82 ± 0.78	79.15 ± 0.74	77.47 ± 0.80
	20	82.09 ± 0.52	80.33 ± 0.53	79.23 ± 0.56	77.46 ± 0.72
	30	79.30 ± 1.02	77.20 ± 1.01	76.21 ± 0.97	74.71 ± 0.84
macro F1	10	47.37 ± 1.55	46.59 ± 1.32	37.43 ± 2.03	43.08 ± 1.19
	20	39.38 ± 1.48	38.23 ± 1.16	25.25 ± 1.66	31.40 ± 1.37
	30	29.56 ± 1.34	28.75 ± 0.89	17.04 ± 0.96	22.13 ± 1.25
micro F1	10	50.75 ± 1.30	47.66 ± 1.35	45.91 ± 2.08	47.90 ± 1.00
	20	44.20 ± 1.22	41.31 ± 1.07	37.33 ± 1.17	40.88 ± 0.89
	30	41.88 ± 1.13	39.45 ± 0.85	34.27 ± 1.19	39.37 ± 0.99
Sensitivity	10	51.49 ± 2.49	60.42 ± 2.95	34.62 ± 2.42	52.04 ± 1.65
	20	39.84 ± 1.76	53.91 ± 1.95	21.28 ± 1.30	33.60 ± 1.38
	30	30.40 ± 1.93	40.27 ± 1.71	14.14 ± 0.83	24.79 ± 1.19
Specificity	10	86.56 ± 0.98	79.57 ± 1.86	92.56 ± 0.55	86.26 ± 0.62
	20	92.07 ± 0.62	85.81 ± 1.11	96.50 ± 0.29	93.33 ± 0.34
	30	94.53 ± 0.46	89.22 ± 0.89	97.73 ± 0.21	95.49 ± 0.28

"ML_LS" denotes the performance obtained by applying the shared subspace multi-label formulation to the proposed bag-of-words representations derived from lateral and dorsal images. "SVM" denotes the performance of SVM applied on the bag-of-words representations using the one-against- rest scheme. "PMK" denotes the method based on pyramid match kernels, and the subscripts "SIFT" and "comp" denote kernels based on the SIFT descriptor and the composite kernels, respectively. Some terms appear in few image groups, and we eliminate them from the experiments. We randomly generate 30 different training/test partitions, and the average performance and standard deviation are reported.

ISSN: 1471-2105