Corrections: Acad Sci USA (106:17355-17358 First Published September 24

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

Corrections

BIOPHYSICS AND COMPUTATIONAL BIOLOGY


Correction for ‘‘A robust method for searching the smallest set
of smallest rings with a path-included distance matrix,’’ by Chang
Joon Lee, Young-Mook Kang, Kwang-Hwi Cho, and Kyoung Tai
No, which appeared in issue 41, October 13, 2009, of Proc Natl
Acad Sci USA (106:17355–17358; first published September 24,
2009; 10.1073/pnas.0813040106).
The authors note that, due to a printer’s error, the Supporting
Information materials did not appear online along with the main
article. The materials have since been made available online. The
online version of the main article has been corrected to include
a link to the Supporting Information.
www.pnas.org/cgi/doi/10.1073/pnas.0911801106

CELL BIOLOGY
Correction for ‘‘CIB1 functions as a Ca2⫹-sensitive modulator of
stress-induced signaling by targeting ASK1,’’ by Kyoung Wan
Yoon, Jun-Ho Cho, Jae Keun Lee, Young-Hee Kang, Ji Soo
Chae, Young Mok Kim, Jeehyun Kim, Eun Kyung Kim, Sung
Eun Kim, Ja-Hyun Baik, Ulhas P. Naik, Ssang-Goo Cho, and
Eui-Ju Choi, which appeared in issue 41, October 13, 2009, of
Proc Natl Acad Sci USA (106:17389–17394; first published
September 29, 2009; 10.1073/pnas.0812259106).
The authors note that during the publication process, the raw
data shown in Fig. 5C were found to be inaccurately labeled and
could not be used to confirm the results. The experiment was
rerun and new data from two repeating experiments were not
significantly different from those originally shown in Fig. 5C,

CORRECTIONS
confirming the results as originally reported. The authors have
submitted a revised figure, replacing the data originally pub- Fig. 5. KCl-elicited Ca2⫹ influx prevents the inhibitory action of CIB1 on
6-OHDA-induced ASK1 activation and apoptosis. (A) SH-SY5Y cells were incu-
lished in Fig. 5C with those of the two new experiments to ensure
bated first with 6-OHDA (100 ␮M) for 50 min and then without or with KCl (40
the raw data are available to readers. The corrected figure and mM) either alone or together with BAPTA-AM (2.5 ␮M) for 10 min, as indi-
its legend appear below. cated. Cell lysates were subjected to immunoprecipitation with anti-ASK1
antibody or rabbit preimmune IgG (P.I.), and the resulting precipitates were
examined by immunoblotting with antibody to CIB1 (Left) or to TRAF2 (Right).
(B) SH-SY5Y cells expressing a control siRNA or CIB1 siRNA were treated with
6-OHDA, KCl, BAPTA-AM as in A. Cell lysates were assayed for ASK1 activity by
immune complex kinase assay. (C) SH-SY5Y cells expressing a control siRNA or
CIB1 siRNA were incubated consecutively with 6-OHDA (100 ␮M) for 50 min,
with KCl (40 mM) either alone or together with BAPTA-AM (2.5 ␮M) for 30
min, and in DMEM containing 2% FBS for 12 h, as indicated. Cells were
analyzed for apoptosis by flow cytometry. The percentages of apoptotic cells
in a representative experiment are indicated.

www.pnas.org/cgi/doi/10.1073/pnas.0911831106

www.pnas.org PNAS 兩 November 17, 2009 兩 vol. 106 兩 no. 46 兩 19629


A robust method for searching the smallest set of
smallest rings with a path-included distance matrix
Chang Joon Leea, Young-Mook Kanga, Kwang-Hwi Chob, and Kyoung Tai Noa,c,1
aDepartment of Biotechnology, Yonsei University, Seoul 120-749, Korea; bDepartment of Bioinformatics and CAMDRC, Soong Sil University,

Seoul 156-743, Korea; and cBioinformatics and Molecular Design Research Center, Seoul 120-749, Korea

Edited by Sung-Hou Kim, University of California, Berkeley, CA, and approved May 1, 2009 (received for review December 22, 2008)

The perception of rings in graphs is widely used in many fields of Table 1. Worst case performance of SSSR algorithm
science and engineering. Algorithms developed in the chemistry Computing Time with Big-Oh notation
community, called smallest set of smallest rings (SSSR), applicable
only for simple graphs or chemical structures. In contrast, algo- With m edges Complete
rithms developed by the computer science community, called Algorithm and n vertices graph*
minimum cycle basis (MCB) are identical to SSSR yet exhibit greater
Horton (11) O(m3n) O(n7)
robustness. MCB-based algorithms can correctly reveal all rings in
De Pina (12) O(m3 ⫹ mn2logn) O(n6 ⫹ n4logn)
any complex graph. However, they are slow when applied to large
Golynski and Horton (14) O(mwn)† O(n2w⫹1)
complex graphs due to the inherent limitations of the algorithms
Berger et al. (15) O(m3 ⫹ mn2logn) O(n6 ⫹ n4logn)
used. Here, we suggest a heuristic method called RP-Path. This Kavitha et al. (16) O(m2n ⫹ mn2logn) O(n5 ⫹ n4logn)
method is a robust, simple, and fast search method with O(n3) Mehlhorn et al. (17) O(m2n2) O(n6)
runtime algorithm that correctly identifies the SSSR of all of the RP-Path (this work) O(n3) O(n3)
test case of complex graphs by using approach different from the
MCB-based method. Both the robustness and improvement in *Worst case, a complete graph, m corresponds to n(n1)/2.
†w ⬍ 2.376.
speed are achieved by using a path-included distance matrix and
describing the characteristic features of rings in the matrix. This
method is accurate and faster than any other methods and may
information and size of the ring simultaneously. With the aids of
find many application in various fields of science and engineering
PID matrices, a robust ring perception method was developed
that use complicated graphs with thousands of nodes.
with a runtime of O(n3). Consisting of two units, ring candidate
graph theory 兩 MCB 兩 ring perception 兩 RP-Path 兩 SSSR
perception followed by ring making and SSSR selection from the
candidates, RP-Path is faster than any other algorithms devel-
oped thus far.
C omplicated systems, for example, biological networks, tele-
communication networks, and electrical circuits are de-
scribed with graphs for mathematical manipulation. Smallest set
RP-Path is consist of two units, (i) ring candidates perception
and (ii) ring making and SSSR selection from the ring
candidates.
of smallest ring (SSSR) is the characteristic feature of such
graphs. Therefore, it is crucial to have robust and fast SSSR Path-Included Distance Matrix. Let G ⫽ (V, E) be a connected
calculation method for accurate analysis and manipulations of a graph with n vertices and m edges. V(G) ⫽ {v1, v2, v3, . . . , vn} is
graph. a finite non-empty set of vertices and E(G) ⫽ {e1, e2, e3, . . . , em}
All reported algorithms developed by the chemistry com- is a finite (possibly empty) set of edges eij represented by
munity to find the SSSR, are based on two main methods, a unordered pairs of vertices in V. The distance matrix D ⫽ D(G)
walk-through connection table (1– 4), and graph manipula- of a graph G with n vertices is the square n⫻ n symmetric matrix
tions (5–8). Some theoretical aspects are summarized by Downs in which diagonal elements are zero, dii ⫽ 0. The element dij
et al. (9). Although various methods to find SSSRs have been corresponds to the minimum path length between vertices i and
developed using algorithms based on both methods, none are j. The Floyd–Warshall method is efficient at finding all-pair
robust (10). shortest paths on a graph (18, 19). For example, the distance
From a computer science perspective, identifying SSSRs, matrix of the graph in Fig. 1 is shown below.
known as minimum cycle basis (MCB), is also interesting. In To find a ring in a circuit with the shortest distance matrix
addition to its SSSR identification capability, the compute time (SDM), Horton’s method uses the distance information between
of a particular method represented by polynomial time is an one point(v) and two points(x and y) connected with one edge,
important consideration toward its performance. Horton pro- C(v, x,y) ⫽ p(v, x) ⫹ p(v, y) ⫹ p(x, y), as described in Fig. 2. Using
posed a robust polynomial time algorithm to find the MCB of a the SDM, this method finds the paths of p(v, x), p(v, y), and p(x, COMPUTATIONAL BIOLOGY

graph (11). The runtime was obtained as O(m3n) for a graph with y). In order for the three points v, x, and y to be members of a
BIOPHYSICS AND

m edges and n vertices. In this approach, the algorithm created ring, the edges of the three paths, p(v, x), p(v, y), and p(x, y), are
a superset of MCB and then extracted it as the shortest m ⫺ n ⫹ not shared among them (Fig. 2 A). In the case of Fig. 2B, the
1 linearly independent cycles from the superset using Gaussian three points are not members of a ring because p(v, x) and p(v, y)
elimination. De Pina (12) proposed a method with a runtime of overlap. Because the SDM contains information on the shortest
O(m3 ⫹ mn2logn) by using a method similar to Padberg and Rao
(13) to solve the minimum weighted T-odd cut problem. Several
methods have been developed based on Horton and De Pina Author contributions: C.J.L., Y.-M.K., K.-H.C., and K.T.N. designed research, performed
methods and their computing performance are summarized in research, contributed new reagents/analytic tools, analyzed data, and wrote the paper.
Table 1 (14–17). The authors declare no conflict of interest.
Here, we present a heuristic method for ring perception This article is a PNAS Direct Submission.
method called RP-Path. Instead of using a distance matrix as in 1To whom correspondence should be addressed. E-mail: ktno@yonsei.ac.kr.
Horton and De Pina’s methods, our method uses path-included This article contains supporting information online at www.pnas.org/cgi/content/full/
distance (PID) matrices, which allow us simply calculate path 0813040106/DCSupplemental.

www.pnas.org兾cgi兾doi兾10.1073兾pnas.0813040106 PNAS 兩 October 13, 2009 兩 vol. 106 兩 no. 41 兩 17355–17358


Table 2. Comparison of Memory requirements of RP-Path and
Horton’s method

Fig. 1. A graph with two SSSRs and its distance matrix.

distances between two vertices only, another time-consuming


procedure is needed to identify the exact path. All possible rings
can be revealed by considering all possible combinations of one
point and two points connected by one edge, C(v, x, y) ⫽ p(v, x) ⫹
p(v, y) ⫹ p(x, y); however, this is also a time-consuming procedure.
To optimize the SSSR identification procedure, variations of Hor-
ton’s method have tried to improve performance by increasing the *Because both methods need distance matrix, D, memory requirement for D
speed of calculating the shortest path between all vertices or by is not included.
eliminating unnecessary calculations for finding rings with the †K is complete graph with n vertices and n is the number of vertices.
n
‡C is cycle graph that consists of a single cycle.
shortest path. However, due to shortcoming inherit in using a n

distance matrix, Horton’s method and its variations have limitations


in reducing computation time.
To overcome the limitations in path search resulting from 兩Pe [i, j]兩 be the number of possible path of Pe [i, j]. If 兩Pe [i, j]兩
using distance matrix, we introduced the PID matrix. The PID is larger than 2 or 兩Pe [i, j]兩 is 1 and 兩P⬘e[i, j]兩 is larger than 1, then
matrix calculation was based on the Floyd–Warshall method, the vertices i and j are members of a ring (Fig. 4).
which was originally developed to calculate the distance matrix. In this manner, if the paths pij,1 and pij,2 in Fig. 4 are the
During the shortest path calculation, the path and the shortest shortest path between vi and vj and the edges are not shared, then
path’s number were stored in the PID matrix, which is crucial for the vertices vi and vj are bounded by two paths pij,1 and pij,2, which
increasing the speed of the ring perception. In addition to form an even-membered ring.
distance matrix, RP-path introduce additional two PID matrices, 兩pij,1兩 ⫽ 兩pij,2兩, pij,1 艚 pij,2 ⫽ 0/
P and P⬘, that store shortest-path information compared with the
other methods in Table 1. RP-path requires additional memory then p ij,1 and p 1j,2 form an even-membered ring [1]
for P and P⬘ than other method. Whereas, the other methods in
Table 1 require more memory than RP-path during ring per- where 兩pij,1兩 and 兩pij,2兩 are the lengths of pij,1 and pij,2, respectively.
ception procedure because other methods have to save all of the If pik is the shortest path (dik) between vi and vk in the Pe matrix
intermediate rings informations(Horton Set). The memory re- and the dij ⫹ 1 path p⬘ik in the P⬘e matrix exists with edges not
quirement varies depending on the types of graphs, a typical shared by the paths pik and p⬘ik, then the vertices vi and vk are
three graph types are shown in Table 2. Each elements of PID bounded by the paths pik and p⬘ik to form an odd-membered ring.
matrix are integers, which are assigned to an edge for an integer,
and have different sizes depending on the length of the path 兩p⬘ik兩 ⫽ 兩pik兩 ⫹ 1, pik 艚 p⬘ik ⫽ 0/
information. For implementation, some programming lan-
guages, e.g., C, might need a indexing to indicate the starting then p ik and p⬘ik form an odd-membered ring [2]
points of each element.
Finding the SSSR. Because a set of all possible rings in a graph {C}
For RP-Path, two matrices, Pe and P⬘e were newly defined. Pe
stores all of the possible shortest paths with a length of dij can be described by two PID matrices Pe and P⬘e, then the next
between vertices i and j. P⬘e stores all paths that have one more step was to select SSSR members {CSSSR} within {C}. The XOR
vertex than the shortest paths with a length of dij ⫹ 1. Pe and P⬘e operation (20) was introduced to select the SSSR from {C}.
for the graph in Fig. 1 are presented in Fig. 3. Once a ring is characterized by the PID matrices, the ring is
With these two matrices, one can identify easily all possible compared with previously identified rings in {CSSSR} by XOR.
rings and the size of these rings. Knowing the size of ring is very If a newly identified ring is not identical to any members of
{CSSSR} or does not contain any members in {CSSSR}, it becomes
important for speeding up in terms of ring perception. Let
a new member of {CSSSR}.
Because large-membered rings may contain smaller rings and
are not members of {CSSSR}, ring identification starts from the
smallest ring (e.g., a three-membered ring) and repeated with
increasing numbers of ring members until the total number of
SSSRs reaches the theoretical number of smallest rings, nSSSR
(nullity, cyclomatic number), as defined by the Euler rule (9).
nSSSR ⫽ m ⫺ n ⫹ 1 [3]

With Horton’s methods, ring size is unknown until all paths are
fully identified. In contrast, RP-Path recognize the ring size and
Fig. 2. The schematic diagram of ring perception in Horton’s method. paths immediately. With this information, RP-Path can start
Examples of a ring made with three points (A) and an incomplete ring (B). with the smallest rings to identify the SSSR. RP-Path consider

17356 兩 www.pnas.org兾cgi兾doi兾10.1073兾pnas.0813040106 Lee et al.


Fig. 3. The PID matrices for the graphs in Fig. 1.

fewer candidate rings than Horton’s methods. A detailed com- The variations of Horton’s methods have tried to focus on
parison of computing time for each stage in Horton’s method increasing the speed of this calculation. However, due to limi-
and RP-Path is summarized in Table 3. tations inherent in using an SDM, Horton’s method and its
variations cannot be improved much more.
Results For the case of a graph with a single cycle, m is equal to n and
The robustness of RP-Path in finding the SSSR was compared the performance of RP-path algorithm is comparable with the
with some widely used SSSR finding methods in chemical De Pina, Berger, and Kavitha algorithms. For the case of a
information, namely the Figueras (21) and Welch (5) methods, complete graph, every pair of distinct vertices are connected by
which are implemented in the chemistry development kit (CDK) edges, m is n(n ⫺ 1)/2 and then m is proportional to n2. In
(22) and PreADMET (BMDRC), respectively. complete graph, RP-Path is faster than the other methods in
A set of graphs tested by Berger et al. was used to test the Table 1. General cases are between the two extreme cases.
robustness of RP-Path in finding SSSR (10). Fig. 5A has a unique
minimum cycle basis consisting entirely of 36 triangles. Our Conclusion
RP-Path identified all three-membered rings correctly, whereas We present a robust and computationally fast method for ring
the Figueras and Welch methods failed to uncover all of perception and SSSR identification from complex graphs. In-
them(Table 4). Moreover, these methods did not produce the stead of using an SDM, RP-Path uses a PID matrix that contains
correct outcome even for simpler test graphs that were analo- information on the paths and size of the rings. With this
gous to the molecular graph, whereas the RP-Path correct information, we developed a robust ring perception method with
identified the SSSR of all graphs (B, C, and D in Table 4). dramatically reduced computing time. This method may have
Computing performance described by the worst case perfor- considerable influence on various fields of science and engi-
mance of RP-Path was compared with Horton’s method (Table neering, including electrical circuit testing (23, 24), periodic
3). RP-Path calculates the PID(Pe and P⬘e) and the distance timetable planning (25), dealing with rigidity of framework
matrices, whereas Horton method calculates only the distance structure (26), chemical ring perception (9), and in network
matrix, Thus, RP-Path has a slightly longer computing time analysis of complicated biological pathways with thousands of
during stage 1. However, the whole procedure for ring percep- nodes.
tion became much simpler and faster once PID was introduced.
In stage 2, because Horton’s method considers three paths using Methods
the SDM, all rings could be identified by considering all possible A ring perception method is presented here followed by the SSSR identifica-
combinations of the three points. In contrast, RP-Path identifies tion method that incorporates all of the features described by the RP-Path
the rings and their size easily with little computation. Because method.
RP-Path discovers rings using a simple algorithm and considers
fewer candidates in finding the SSSR, The computing speed of
Table 3. Comparison of worst case performance for each stage
the stages 3 and 4 are much faster than that of Horton’s method.
Horton RP-Path (this work)

Stage Method BigOh Method BigOh

1 Distance Matrix O(n3) Distance matrix, O(n3)* COMPUTATIONAL BIOLOGY

Calculation PID matrices(P,


BIOPHYSICS AND

P⬘) calculation

2 Ring Search O(mn2)† Ring candidate O(n2)


search
3 Sort({C}) O(a log a)‡ Sort (Cset) O(a⬘ log a⬘)§
4 Finding O(m3n) Finding SSSR O(a⬘k)
MCB(SSSR)

*n is the number of vertices.


†m is the number of edges.
‡a is the total number of circle({C}).
§a⬘ is the total number of ring candidate({C }).
set
Fig. 4. A path diagram between vertices. ¶k is the number of the minimum path between i and j.

Lee et al. PNAS 兩 October 13, 2009 兩 vol. 106 兩 no. 41 兩 17357
Table 4. Comparison of ring perception results for the graphs
in Fig. 5
Number of n-membered ring
n-membered
Graph ring/Nurity Figureas (21) Welch (5) This work

3*/36 25† 32 36
4/0 8 1 0
A 5/0 1 1 0
7/0 2 1 0
8/0 0 1 0
3/0 0 0 0
B 4/6 5 6 6
5/2 1 2 2
6/0 2 0 0
3/12 10 12 12
C 4/0 1 0 0
5/2 2 1 2
6/0 1 1 0
3/5 5 5 5
D 4/1 1 1 1
6/1 1 0 1
7/0 0 1 0

Fig. 5. Graphs introduced by Berger et al. for testing the robustness of the
SSSR finding algorithm. *The number of x-membered rings.
†25 means 25 of 3-membered rings.

Y Step 1. Take information from a graph with M vertices and N edges.


Y Step 7. Construct ring using the PID matrix from the smallest-membered
Y Step 2. Remove all open acyclic nodes with one edge(connectivity equal to
rings.
1, corresponding to terminal nodes) and repeat the procedure, stoping at
Y Step 8. Use the XOR operation to find the SSSR from this ring.
nodes with a connectivity larger than 1. Now, M and N become m and n,
Y Step 9. If a ring is a new member of the SSSR, then save it.
respectively.
Y Step 10. Repeat steps 7 to 9, until nringidx reaches nSSSR, where nringidx is the
Y Step 3. Calculate the theoretical number of SSSR : nSSSR ⫽ m ⫺ n ⫹ 1, If nSSSR ⫽
number of founded SSSR.
0, stop process.
Y Step 4. Calculate the PID matrices Pe and P⬘e.
ACKNOWLEDGMENTS. This work was partly supported by the IT R&D program
Y Step 5. Create the cycle set([Cnum, Pe[i,j], P⬘e[i,j]]) using the PID matrices. Cases
of MKE/IITA, which is funded by MKE/IITA Grant 2008-F-029-01, Development
in which the shortest path number equals zero, or [兩Pe[i,j]兩 ⫽ 1 and 兩P⬘e[i,j] ⫽ of e-Organ system based on Cyber Computing; and the 21st Century Frontier
0] can be omitted. R&D Program Grant CBM31-B2000-01-00-00 of the Center for Biological Mod-
Y Step 6. Order the cycle set by Cnum number. ulators funded by the Ministry of Education, Science and Technology, Korea.

1. Bersohn M (1973) An algorithm for finding the synthetically important rings of a 14. Golynski, A. and Horton, J (2002) A polynomial time algorithm to find the minimum
molecule. J Chem Soc Perkin Trans 1 15:1239 –1241. cycle basis of a regular matroid. Lecture Notes Comput Sci 200 –209.
2. Esack A (1975) A procedure for rapid recognition of the rings of a molecule. J Chem Soc 15. Berger F, Gritzmann P, de Vries S (2004) Minimum cycle bases for network graphs.
Perkin Trans 1 4:1120 –1124. Algorithmica 40:51– 62.
3. Zamora A (1976) An algorithm for finding the smallest set of smallest rings. J Chem Inf 16. Kavitha T, Mehlhorn K, Michail D, Paluch K (2004) A faster algorithm for minimum cycle
Comput Sci 16:40 – 43. basis of graphs. Lecture Notes in Comput Sci 846 – 857.
4. Corey E, Perersson G (1972) Algorithm for machine perception of synthetically signif- 17. Mehlhorn K, Michail D (2005) Implementing minimum cycle basis algorithms. In
icant rings in complex cyclic organic structures. J Am Chem Soc 94:460 – 465. Proceedings of the Fourth International Workshop on Efficient and Experimental
5. Welch J, Jr (1966) A mechanical analysis of the cyclic structure of undirected linear Algorithms (WEA) (Springer, Heidelberg).
graphs. J ACM 13:205–210. 18. Floyd R (1962) Algorithm 97: Shortest path. Commun ACM 5:345.
6. GIBBS, N (1969) A cycle generation algorithm for finite undirected linear graphs. J ACM
19. Warshall S (1962) A theorem on boolean matrices. J ACM 9:11–12.
16:564 –568.
20. Wipke W, Dyott T (1975) Use of ring assemblies in ring perception algorithm. J Chem
7. Fujita S (1988) Logical perception of ring-opening, ring-closure, and rearrangement
Inf Comput Sci 15:140 –147.
reactions based on imaginary transition structures. Selection of the essential set of
21. Figueras J (1996) Ring perception using breadth-first search. J Chem Inf Comput Sci
essential rings (ESER). J Chem Inf Comput Sci 28:1–9.
36:986 –991.
8. Fujita S (1988) A new algorithm for selection of synthetically important rings. The
22. Steinbeck C, et al. (2003) The chemistry development kit (CDK): An Open-source Java
essential set of essential rings for organic structures. J Chem Inf Comput Sci 28:78 – 82.
9. Downs G, Gillet V, Holliday J, Lynch M (1989) Review of ring perception algorithms for library for chemo- and bioinformatics. J Chem Inf Comput Sci 43:493–500.
chemical graphs. J Chem Inf Comput Sci 29:172–187. 23. Chua L, Chen L (1973) On optimally sparse cycle and coboundary basis for a linear
10. Berger F, Flamm C, Gleiss P, Leydold J, Stadler P (2004) Counterexamples in chemical graph. IEEE Trans Circuits Syst I Regul Pap 20:495–503.
ring perception. J Chem Inf Comput Sci 44:323–331. 24. Brambilla A, Premoli A, Inf D (2001)Rigorous event-driven (RED) analysis of large-scale
11. Horton J (1987) A polynomial-time algorithm to find the shortest cycle basis of a graph. nonlinear RCcircuits. IEEE Trans Circuits Syst I Regul Pap 48:938 –946.
SIAM J Comput 16:358. 25. Liebchen C, Misohring R (2002) A case study in periodic timetabling electron. Notes
12. de Pina J (1995) in Applications of Shortest Path Methods (University of Amsterdam, Theor Comput Sci 66:18 –31.
Amsterdam). 26. Spiliopoulos K (1997) On the automation of the force method in the optimal plastic
13. Padberg M, Rao M (1982) Odd minimum cut-sets and b-matchings. Math Oper Res 7:67– 80. design of frames. Comput Methods Appl Mech Engrg 141:141–156.

17358 兩 www.pnas.org兾cgi兾doi兾10.1073兾pnas.0813040106 Lee et al.

You might also like