Figure 4

Schematic description of the discretization protocols used. On the left is shown the original matrix containing continuous values expressed as copies of transcript per cell, in the center the discretization procedure used, and on the right the resulting boolean matrix. (a) The max minus 25% approach. We took the highest value (65) and calculated 25% of 65 (16.25). Values above 65 - 16.25 (= 48.75) were given a boolean value of 1; all others were given a value of 0. (b) The mid-range approach. We took the highest value (65), the lowest value (18), and calculated the mid-range as ((65-18)/2)+18= 41.5. Values above 41.5 were given a boolean value of 1; all others were given a value of 0. (c) The 30% cut-off. We took the highest 30% of the number of values (here 7 × 0.3 = 2.1, that is, the two highest values), set them to 1 and the rest to 0. Note that this 30% value is not the actual value used for the extraction (see below) but is given as an illustration. Note also that these three different procedures resulted in different matrix densities. The last procedure resulted in the same number of boolean 1 results for all tags, whereas the first two generated different densities of 1, depending on the gene-expression pattern throughout the various libraries.