Download abcedf

B
A
Number of complexes
Functional homogentity in PI complexes
3.5
P < 0.001
3
2.5
2
100
R = 0.878
10
1
0.1
2
1
10
100
Number of subunits
1.5
C
1
Distribution of node connectivity
1000
Integrated PI score >= 0.7
0.5
Number of nodes
Cluster cohesiveness (Z-score)
Distribution of complex size
1000
optimal MCL
inflation
Integrated PI score >= 0.75
mcl1.2
mcl2
mcl3
mcl4
mcl5
mcl1.2
mcl2
mcl2.1
mcl3
mcl4
mcl5
mcl1.2
mcl2
mcl3
mcl4
mcl5
mcl1.2
mcl2
mcl3
mcl4
mcl5
mcl1.2
mcl2
mcl3
mcl4
mcl5
mcl1.2
mcl2
mcl3
mcl4
mcl5
mcl1.2
mcl2
mcl3
mcl4
mcl5
0
0 hubs
10 hubs
20 hubs
30 hubs
50 hubs
100 hubs
100
2
R = 0.906
10
1
200 hubs
0.1
0.1
1
10
100
Number of edges
E
D
1000
Number of modules
40
35
30
P << 0.001
25
20
100
R 2 = 0.783
10
1
0.1
1
15
10
100
Number of subunits
F
10
Distribution of node connectivity
Integrated GC score >= 0.75
5
optimal MCL
inflation
Integrated GC score >= 0.8
1000
0 hubs
10 hubs
100 hubs
mcl5
mcl4
mcl3
mcl2
mcl1.2
mcl5
mcl4
mcl3
mcl2
mcl1.2
mcl5
mcl4
mcl3
mcl2
mcl1.7
mcl1.2
0
Number of nodes
Cluster cohesiveness (Z-score)
Distribution of module size
Functional homogentity in GC modules
R 2 = 0.783
100
10
1
0.1
0.1
1
10
100
100
Number of edges
Figure S2 – Functional homogeneity, connectivity and cluster size in PI complexes and GC modules.
Three parameters were tuned to determine the optimal structural and functional partitioning of networks by MCL into PI multiprotein complexes and
GC functional modules: namely, the MCL inflation value, the interaction scores, and the number of highly-connected hubs removed from consideration
from each network. Optimization of these parameters resulted in a significantly greater functional homogeneity as compared with null models for both
the PI and GC networks, as shown in panels (A) and (D), respectively. The distribution of members in the resulting final optimized sets of
(B) putative complexes and (E) functional modules fits a power-law distribution. Conversely, while the degree of connectivity of the high-confidence PI
network (C) follows a typical power-law distribution, the degree of connectivity of the GC network (F) shows an increased number of nodes with 10 to
100 interactions, deviating from a power law distribution.