Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Dynamical robustness of the Mammalian cell cycle network: a mathematical approach Marco Montalva, Eric Goles and Gonzalo A. Ruz Facultad de Ingenierı́a y Ciencias, Universidad Adolfo Ibáñez, Santiago, Chile ECCS’13 - Barcelona, September 16-20, 2013 Summary Fixed points in Case 1 (Main Theorem) A logical model for the control of the mammalian cell cycle was presented in [1] where the simulations performed to obtain the dynamic of the network, in parallel mode, showed that in the presence of Cyclin D protein (CycD), any trajectory converges to a unique limit cycle. In this work, we prove mathematically that, in fact, the presence of CycD is a necessary and sufficient condition to assure that the network dynamics has no fixed point, regardless of the deterministic updating mode considered (in particular, the parallel one). Finally, we present a detailed study of all the deterministic dynamics including the other case (absence of CycD). Mammalian cell cycle model The logical model for the control of the Mammalian cell cycle [1,3] CycD CycE p27 Rb E2F CycA Cdc20 UbcH10 Cdh1 CycB Theorem: Let S be the set of all the deterministic updating schemes of the Mammalian network. Then, CycD = 1 ⇔ ∀s ∈ S, the mammalian dynamics, generated by the nine remaining nodes, has no fixed point. Proof: ⇐) If CycD = 0, then by [1], the Mammalian dynamic has one fixed point for the parallel update. ⇒) Let CycD = 1 and s ∈ S. We know that in at most one iteration step, Rb = 0 and p27 = 0. Therefore, the Mammalian dynamic generated by the seven remaining nodes can be described by the dynamic of the above subnetwork. Let N be such subnetwork, we will prove (by contradiction) that N has no fixed point. Let us consider that y 0 = (E2F, CycE, CycA, Cdc20, Cdh1, UbcH10, CycB) ∈ {0, 1}7 is a fixed point of N. We can consider two possibilities; Cdc20 = 1 or Cdc20 = 0 and we will have the following implications for each: I If Cdc20 = 1, then CycB = 1 in (14) but, also, if Cdc20 = 1, then CycB = 0 in (17), which is a contradiction. I If Cdc20 = 0, then CycB = 0 in (14). If (Cdc20 = 0 ∧ CycB = 0), then Cdh1 = 1 in (17). If (Cdc20 = 0 ∧ CycB = 0 ∧ Cdh1 = 1), then CycA = 0 in (15). If (Cdc20 = 0 ∧ Cdh1 = 1 ∧ CycA = 0), then (E2F = 0 ∨ UbcH10 = 1) in (13) but, on the other hand, if we consider the above values of; CycA = 0, CycB = 0, Cdc20 = 0 and Cdh1 = 1, we will have that E2F = 1 and UbcH10 = 0 in (11) and (16) respectively, which is a contradiction. Therefore, N cannot have fixed points. Local boolean functions Mammalian cell cycle network dynamic with the following updating scheme: (CycD, Rb, Cdc20, Cdh1, CycA)(p27, UbcH10, CycB)(E2F)(CycE). The red circle represents the fixed point, the 12 green circles represent the states that belong to two limit cycles, one of length 8 and another of length 4: Pajek Example 2: Dynamic with 3 limit cycles Mammalian cell cycle network dynamic with the following updating scheme: (CycD, p27, Cdc20, Cdh1, UbcH10, CycB)(E2F)(CycE)(Rb, CycA). The red circle represents the fixed point, the 14 green circles represent the states that belong to three limit cycles, two of length 6 and one of length 2: Dynamics summary for Case 1 (1) Rb =(CycD ∧ CycB) ∧ ([CycE ∧ CycA] ∨ p27) (2) E2F =(Rb ∧ CycA ∧ CycB) ∨ (p27 ∧ Rb ∧ CycB) (3) CycE =(E2F ∧ Rb) (4) CycA =(Rb ∧ Cdc20 ∧ Cdh1 ∧ UbcH10) ∧ (E2F ∨ CycA) (5) p27 =(CycD ∧ CycB) ∧ ([CycE ∧ CycA] ∨ [p27 ∧ CycE ∧ CycA]) (6) Cdc20 =CycB (7) Cdh1 =(CycA ∧ CycB) ∨ Cdc20 ∨ (p27 ∧ CycB) (8) UbcH10 =Cdh1 ∨ (Cdh1 ∧ UbcH10 ∧ [Cdc20 ∨ CycA ∨ CycB]) (9) CycB =Cdc20 ∧ Cdh1 (10) CycD =CycD Our analysis is separated in two cases; CycD = 1 and CycD = 0. Different dynamics 1696 |D2| |D3| |D4| |D5| |D6| |D7| 16 298 812 432 132 6 (0.9%) (17.6%) (47.9%) (25.5%) (7.8%) (0.4%) Attractor analysis for all the dynamics of the subnetwork of seven nodes. In this case, there is only one limit cycle by dynamic, with attraction basin of size 512 (including the limit Some previous simplifications: I If CycD = 1, then Rb = 0 in (2) and p27 = 0 in (6). I If (Rb = 0 ∧ p27 = 0), then E2F = CycA ∧ CycB in (3). I If Rb = 0, then CycE = E2F in (4). I If Rb = 0, then CycA = (Cdc20 ∧ Cdh1 ∧ UbcH10) ∧ (E2F ∨ CycA) in (5). I If p27 = 0, then Cdh1 = (CycA ∧ CycB) ∨ Cdc20 in (8). i.e., the Mammalian network can be reduced to the following subnetwork: CycE E2F CycA only limit cycle has length i. We are use the algorithm I We are presented a mathematical approach for the Mammalian network [1] that consist in to separate its The following table show the general attractor analysis for all the dynamics of the subnetwork of nine nodes obtained when CycD = 0: CycE =E2F CycA =(Cdc20 ∧ Cdh1 ∧ UbcH10) ∧ (E2F ∨ CycA) Cdc20 =CycB Cdh1 =(CycA ∧ CycB) ∨ Cdc20 UbcH10 =Cdh1 ∨ (Cdh1 ∧ UbcH10 ∧ [Cdc20 ∨ CycA ∨ CycB]) be divided into two components; one containing half of the Different 130297 2-8 CycD = 1), the other one, with a single fixed point and, with 1 lim. cycle 19719 (15.1%) 2-8 possibly, with at most two limit cycles (case CycD = 0). with 2 lim. cycles 163 (0.1%) 2-6 without lim. cycles 110415 (84.7%) – I Note that most of the dynamics (≈ 85%) have a single attractor; a fixed point. I In the rest of the dynamics (≈ 15%), only one limit cycle appears except for a few ones (≈ 0.1%) where two additional limits cycles appears. A more detailed analysis for the dynamics having a limit cycle is showed in the next table: Cdh1 (11) (12) (13) (14) (15) (16) (17) I Every deterministic dynamic of the Mammalian network can Range of lim. cycles length UbcH10 E2F =CycA ∧ CycB dynamic according to the possible values of node CycD. Number of dynamics Dynamics with Average at least 1 Basin Size and, with the following local functions on its nodes: Conclusion developed in [2] for the classification of identical dynamics. Cdc20 CycB Pajek cycle configurations). |Di| is the number of dynamics whose Case 2: CycD = 0 Case 1: CycD = 1 CycB =Cdc20 ∧ Cdh1 Example 1: Dynamic with 2 limit cycles lim. cycle of length 2 6587 140.5 lim. cycle of length 3 3383 321.7 lim. cycle of length 4 6148 273.4 lim. cycle of length 5 2892 298.1 lim. cycle of length 6 919 325 lim. cycle of length 7 21 363.3 lim. cycle of length 8 63 372.9 I The most frequent limit cycle lengths are 2 and 4 whereas the least frequent are 7 and 8, however, these latest have the maximum average basin sizes whereas the first ones have the minimum average basin sizes configurations converging to a single limit cycle (case I When CycD = 1, the range of limit cycles lengths is between 2 and 7, being 4 the most frequent and with basin size of 29 = 512. I When CycD = 0, the maximum length of limit cycles grows up 8, being the most frequent the length 2 (but with average basin size less than other longer limit cycles). References [1] A. Fauré, A. Naldi, C. Chaouiya, D. Thieffry, Dynamical analysis of a generic boolean model for the control of the mammalian cell cycle, Bioinformatics 22 (14) (2006), 124–131. [2] J. Aracena, J. Demongeot, E. Fanchon, M. Montalva, On the number of different dynamics in boolean networks with deterministic update schedules, Mathematical Biosciences 242 (2) (2013), 188–194. [3] G. A. Ruz, E. Goles, Reconstruction and update robustness of the mammalian cell cycle network, in: The 2012 IEEE Symposium on Computational Intelligence in Bioinformatics and Computational Biology (CIBCB 2012), San Diego, California, USA, 2012, 397–403. Acknowledgment: Fondecyt 3130466 (M.M.), Fondecyt 1100003 (E.G.), Fondecyt 11110088 (G.A.R.), Basal (Conicyt)-CMM and Anillo Act-88