Table 2 The 60 binary classifiers used in the ensLOC framework
Classifier IDName of Binary ClassifierNo. of Positive Training ObjectsNo. of Negative Training Objects-Validation Using 10-fold Cross-ValidationVisual Inspection Recall
RecallPrecision
Quality control
 1.1.1DEAD96015410.9860.995
 1.1.2GHOST184023980.9951
Budded or Unbudded
 2.1.1UNBUDDED109515820.9970.984
 2.1.2SMALLBUDDED4347330.9520.948
 2.1.3LARGEMEDIUMBUDDED72715080.9850.986
3.1 Cytoplasm
 3.1.1CYTOPLASM349342850.9790.966∼95%
 3.1.2CYTOPLASMNOTNUCLEAR207514190.9150.842>95%
3.2 Endosome
 3.2.1ENDOSOME224547300.8260.912<70%
 3.2.2ENDOSOME_CYTOPLASM224534930.9770.995
 3.2.3ENDOSOME_NUCLEI224556120.9950.999
 3.2.4ENDOSOME_SPINDLEPOLE224533970.9630.986
 3.2.5ENDOSOME_MITOCHONDRIA224563150.8990.967
3.3 ER
 3.3.1ER527442590.9770.919<80%
 3.3.2ER_CYTOPLASM527434930.970.965
 3.3.3ER_VACUOLEVACUOLARMEMBRANE527438930.9760.958
 3.3.4ER_CELLPERIPHERY527440590.9960.996
3.4 Golgi
 3.4.1GOLGI199418380.9640.908>80%
 3.4.2GOLGI_MITOCHONDRIA199463150.8090.968
 3.4.3GOLGI_ENDOSOME199422450.9190.934
 3.4.4GOLGI_CYTOPLASM199434930.9960.999
3.5 Mitochondria
 3.5.1MITOCHONDRIA631578940.8940.884>85%
3.6 Nuclear Periphery
 3.6.1NUCLEARPERIPHERY266843670.940.96∼70%
3.7 Nucleus
 3.7.1NUCLEI561268810.9770.956>80%
 3.7.2NUCLEINOTCYTOPLASM13989890.990.93>80%
3.8 Nucleolus
 3.8.1NUCLEOLUS388253320.9260.948>85%
3.9 Peroxisome
 3.9.1PEROXISOME125620990.8490.922<70%
 3.9.2PEROXISOME_GOLGI125619930.9280.971
 3.9.3PEROXISOME_SPINDLEPOLE125633970.9650.995
 3.9.4PEROXISOME_MITOCHONDRIA125663150.8140.981
3.10 Vacuole/Vacuolar Membrane
 3.10.1VACUOLEVACUOLARMEMBRANE-COMBINED389333520.9260.898>80%
 3.10.2VACUOLE_VACUOLARMEMBRANE222418460.920.845>80% VAC, 65% VAC membrane
3.11 Cortical Patches
 3.11.1CORTICALPATCHESUNBUDDED181312790.9640.877∼70%
 3.11.2CORTICALPATCHESUNBUDDED_CYTOPLASM181316610.9940.996
 3.11.3CORTICALPATCHESUNBUDDED_MITOCHONDRIA181344400.950.984
 3.11.4CORTICALPATCHESBUDDED134521710.9280.93675%
 3.11.5CORTICALPATCHESBUDDED_CELLPERIPHERY134510590.9940.988
 3.11.6CORTICALPATCHESBUDDED_MITOCHONDRIA134518750.9810.986
 3.11.7CORTICALPATCHESBUDDED_CYTOPLASM134510220.9870.988
3.12 Bud
 3.12.1BUD161916910.9370.905>70%
3.13 Budneck
 3.13.1BUDNECK217030950.9470.942>70%
 3.13.2BUDNECK_BUD217016190.9620.946
 3.13.3BUDNECK_CELLPERIPHERY2170105910.994
 3.13.4BUDNECK_MITOCHONDRIA217018750.990.98
 3.13.5BUDNECK_CYTOPLASM217010220.9870.98
 3.13.6BUDNECK_NUCLEI2170131310.996
3.14 Budsite
 3.14.1BUDSITE4536370.9820.961>80%
 3.14.2BUDSITE_CYTOPLASM45349550.9430.992
 3.14.3BUDSITE_CELLPERIPHERY4533590.9960.992
3.15 Cell Periphery
 3.15.1CELLPERIPHERYUNBUDDED22698580.9890.98>95%
 3.15.2CELLPERIPHERYBUDDED105916880.9810.991>85%
3.16 Spindle Pole
 3.16.1SPINDLEPOLETWODOTFARBUDDED4169660.9380.965>70%
 3.16.2SPINDLEPOLETWODOTFARBUDDED_BUDNECK41621700.9130.997
 3.16.3SPINDLEPOLETWODOTFARBUDDED_NUCLEARPERIPHERY41649210.996
 3.16.4SPINDLEPOLETWODOTFARBUDDED_NUCLEOLUS41611090.990.995
 3.16.5SPINDLEPOLETWODOTCLOSEBUDDED30610160.9050.97∼80%
 3.16.6SPINDLEPOLETWODOTCLOSEBUDDED_BUDNECK30621700.8990.995
 3.16.7SPINDLEPOLETWODOTCLOSEBUDDED_MITOCHONDRIA30618750.9740.996
 3.16.8SPINDLEPOLETWODOTCLOSEBUDDED_NUCLEARPERIPHERY3064920.9930.996
 3.16.9SPINDLEPOLETWODOTCLOSEBUDDED_NUCLEOLUS30611090.980.988
 3.16.10SPINDLEPOLEONEDOT267536760.9740.98370%
  • In total, approximately 70K handpicked cell images (objects) were used to train the classifiers. “No. of positive training objects” refers to cells which belong to the targeted class and “No. of negative training objects” refer to cells not belonging to the targeted class. For example, to construct the “DEAD” cells classifier, 960 images of dead cells were used as positive training objects and 1541 images of non-dead cells from across all 16 localization classes were used as negative training objects. The first number of the classifier ID reflects the level and therefore the sequence at which the classifier was applied. For instance, all cell images were first tested using the “DEAD” cells classifier to eliminate dead cells from further classification to the 16 localization classes, and only cells that were tested positive in the level 2 “SMALLBUDDED” and “LARGEMEDIUMBUDDED” classifiers would be further classified by the “BUDNECK” classifier. The accuracy of the classifiers was validated computationally using 10-fold cross-validation and manually using visual inspection of 500 random positive cells. Recall = True positives/(True positives + False negatives); Precision = True positives/(True positives + False positives). ER, endoplasm reticulum.