A machine learning approach for genome-wide prediction of morbid and druggable human genes based on systems-level data

Table 2 Statistical comparison of performances of classifiers trained on normal and without-one-feature morbidity datasets

Missing feature ¹	Median AUC [min,max]²	N	W	W_c (two-tailed p = 0.05)³
ppi	0.715 [0.705,0.726]	10	26	8
metin	0.714 [0.707,0.727]	10	26	8
metout	0.713 [0.707,0.729]	10	25	8
regin	0.714 [0.703,0.726]	9	18	6
regout	0.716 [0.705,0.729]	10	26	10
c	0.713 [0.701,0.724]	10	13	8
identicalness	0.711 [0.704,0.727]	10	24	8
cent	0.714 [0.707,0.727]	10	25	8
inbet	0.716 [0.708,0.731]	10	25	8
inbetppi	0.714 [0.707,0.727]	9	21	6
inbetmet	0.714 [0.707,0.728]	9	21	6
inbetreg	0.715 [0.706,0.727]	10	25	8
numtissuesexp ⁴	0.709 [0.701,0.719]	10	7	8*
avegexptec ⁵	0.715 [0.704,0.727]	10	27	8
Unknown	0.713 [0.701,0.725]	10	18	8
Cytoplasm	0.715 [0.706,0.728]	10	26	8
Endoplasmic reticulum	0.716 [0.705,0.727]	10	26	8
Mitochondrion	0.714 [0.706,0.728]	10	24	8
Nucleus	0.715 [0.704,0.728]	10	24	8
Other localization	0.714 [0.704,0.726]	10	21	8
Cellular component	0.714 [0.705,0.727]	9	21	6
Extracellular space	0.710 [0.7,0.723]	10	14	8
Golgi apparatus	0.715 [0.706,0.728]	10	26	8
Median AUC [min,max] for normal datasets: 0.716 [0.706,0.729]

¹ See “Methods” and Additional file 1 for a description of features
² Of 10 datasets
³ According to table of critical values for W in [6]
⁴ The number of tissues (out of 32) in which the gene is expressed at least 5 transcripts per million (tpm) according to Reverter et al. [33]
⁵ The average expression in tpm among all the tissues in which the gene is expressed according to Reverter et al. [33]
* Difference statistically significant

ISSN: 1471-2164