Figure 1.
Cells from different organisms are clustered separately in the lineage tree.
Two reconstructed lineage trees are shown: (A) A lineage tree containing cells from three mice, M3 (blue), M5 (pink) and M6 (green). (B) A lineage tree containing cells from seven humans, H1 (blue), H2 (red), H3 (orange), H4 (pink), H5 (green), H6 (purple) and H7 (turquoise). The root of the trees (colored in black) is the weighted mean of the organisms' putative zygotes. The trees were reconstructed using the NJ algorithm along with the Normalized-Absolute distance measure.
Figure 2.
Cells from different organisms are mixed in the lineage tree.
A reconstructed lineage tree containing cells from six mice, M1 (turquoise), M2 (red), M3 (blue), M7 (yellow), and M8 (purple). The root of the trees (colored in black) is the weighted mean of the organisms' putative zygotes. The trees are reconstructed using the NJ algorithm along with the Normalized-Absolute distance measure.
Figure 3.
Performance of different methods on the same dataset.
Two reconstructed lineage trees containing the same cells from three different mice are shown: M1 (turquoise), M3 (blue), and M4 (orange). (A) A tree using the NJ algorithm along with the Normalized-Absolute distance measure. (B) A tree using the NJ algorithm along with the Equal or Not distance measure. The root of the trees (colored in black) is the weighted mean of the organisms' putative zygotes. It can be seen that the performance of the Normalized-Absolute distance measure is clearly better.
Table 1.
Example of results on a few datasets.
Figure 4.
Performance summary of all the methods on all the datasets of mice and humans.
Upper panels – Mouse, Lower panels- Human. Each column presents a different clustering measure (see Materials and Methods for details), and each bar represents a different distance measure, where the colors specify the distance measures as noted in the legend. The first and the second group of bars (from left to right) present the results using the NJ algorithm and the QMC algorithm respectively. Rows description: (A) The average score of all the methods, where higher values (that are transformations of the real scores) indicate better performance. (B) The number of times every method received the highest rank (for the mouse panel the highest rank is 20 since we compared 20 methods, and for the human panel the highest rank is 12).
Figure 5.
Performance summary of all the methods for datasets composed of cells of same depth only.
Each column presents a different clustering measure (see Materials and Methods for details), and each bar represents a different distance measure, where the colors specify the distance measures as noted in the legend. The first group of bars (from left to right) presents the results using the NJ algorithm, the second group of bars presents the results using the QMC algorithm, the third presents the results using the UPGMA algorithm, and the last one presents the results using the BATWING tool. Rows description: (A) The average score of all the methods, where higher values (that are transformations of the real scores) indicate better performance. (B) The number of times every method received the highest rank (in this case it is 31 since we compared 31 methods).
Figure 6.
Depth separation comparison of two distance measures on the same dataset.
Two reconstructed lineage trees containing whole crypts from M9 (52 days) and two reconstructed lineage trees containing whole crypts from M7 (199 days) are shown. The root of the trees (colored in black) is the signature of the tail extracted from each mouse. (A) Two reconstructed trees using the NJ algorithm along with Equal or Not distance measure. (B) Two reconstructed trees using the NJ algorithm along with the Absolute distance measure.
Figure 7.
Depth quality of the different methods.
Each column presents a different depth quality measure (see Materials and Methods for details), and each bar represents a different distance measure, where the colors specify the distance measures as noted in the legend. Rows description: (A) The average depth score of all the methods, where lower values for the KS test, lower values for the overlap percentage and higher values for the normalized distance test mean better performance. (B)The number of times every method received the highest rank (in this case it is 10 since we compared 10 methods).