A COMPARATIVE ANALYSIS OF PROGRESSIVE MULTIPLE SEQUENCE ALIGNMENT APPROACHES USING UPGMA AND NEIGHBOR JOIN BASED GUIDE TREES

International Journal of Computer Science, Engineering and Information Technology (IJCSEIT), Vol. 5,No.3/4, August 2015
DOI : 10.5121/ijcseit.2015.5401 1
A COMPARATIVE ANALYSIS OF PROGRESSIVE
MULTIPLE SEQUENCE ALIGNMENT APPROACHES
USING UPGMA AND NEIGHBOR JOIN BASED GUIDE
TREES
Dega Ravi Kumar Yadav1
and Gunes Ercal2
Department of Computer Science, SIUE, Edwardsville, Illinois, USA
ABSTRACT
Multiple sequence alignment is increasingly important to bioinformatics, with several applications ranging
from phylogenetic analyses to domain identification. There are several ways to perform multiple sequence
alignment, an important way of which is the progressive alignment approach studied in this work.
Progressive alignment involves three steps: find the distance between each pair of sequences; construct a
guide tree based on the distance matrix; finally based on the guide tree align sequences using the concept
of aligned profiles. Our contribution is in comparing two main methods of guide tree construction in terms
of both efficiency and accuracy of the overall alignment: UPGMA and Neighbor Join methods. Our
experimental results indicate that the Neighbor Join method is both more efficient in terms of performance
and more accurate in terms of overall cost minimization.
KEYWORDS
Progressive MSA, Guide Tree, Profiles, Pair-wise Distance & Dynamic Programming.
1. INTRODUCTION
The traditional pairwise sequence alignment problem in its utmost generality is to find an
arrangement of two given strings, S and T, such that the arrangement yields information on the
relationship between S and T, such as the minimum number of changes to S that would transform
S into T. In the context of DNA sequences, which can be viewed as strings from the 4 letter
alphabet {A, C, G, T}, these changes may represent mutation events, so that the alignment sought
yields important evolutionary information [15]. Similarly, the pairwise sequence alignment
problem can be generalized to the multiple sequence alignment problem to yield information on
the relatedness of multiple sequences. Applications of the multiple sequence alignment (MSA)
problem for DNA sequences include phylogenetic analysis, domain identification, discovery of
DNA regulatory elements, and pattern identification. Additionally, MSA applications for protein
sequences also includes protein family identification and structure prediction. This work is
concerned with approaches to multiple sequence alignment in the context of DNA sequences.
Generally, aligning two sequences is straightforward via dynamic programming. But pairwise
alignment is insufficient for many applications in which the relationship among several sequences
is sought. Moreover, it is infeasible to naturally extend the dynamic programming approach that
works for pairwise sequence alignment directly to multiple sequence alignment when there are
more than three sequences to align. Unfortunately, multiple sequences alignment is NP-hard
based on SP (sum-of-pairs) scores [1]. Therefore, heuristics are crucial to MSA.

International Journal of Computer Science,
Progressive alignments are by far
method [2, 3]. Progressive alignment
calculations for all the input DNA
computed in the previous step. 3.
nearest sequences first. In this
approaches to MSA in terms of
progressive MSA with UPGMA
based guide trees. Our results indicate
preferable in terms of both efficiency
2. BACKGROUND AND RELATED
The main steps of the progressive
1. Compute pairwise distances
2. Build the guide tree based
3. Align first two sequences
(Global Alignment).
4. From the next sequence
profile.
5. Repeat step 4 until the longest
6.
Figure 1:
2.1. Pairwise Distance
The distance between two DNA
distance matrix is computed among
number of evolutionary models
Jukes-Cantor model [8]. The Jukes
matches to the number of non-gaps
distance formula is:
Here `D' is the Jukes-Cantor distance
of matches to the number of non
specified species, a guide tree based
2.2. Guide Tree
In progressive alignment, the guide
determine which sequence is to be
a phylogenetic tree that is constructed
A phylogenetic tree is an evolutionary
species. Phylogenetic trees are
Science, Engineering and Information Technology (IJCSEIT), Vol. 5,No.3/4,
far the most widely used heuristic multiple sequence
alignment is done in three major steps; 1. Perform pairwise
DNA sequences. 2. Build the guide tree using the distance
3. Based on the guide tree, perform progressive alignment
this project, we compare two important progressive
of algorithmic efficiency as well as alignment accuracy,
UPGMA based guide trees and progressive MSA with Neighbor
indicate that the Neighbor Join method of guide tree construction
efficiency and accuracy of the overall resulting MSA.
RELATED WORK
progressive alignment methodology are as follows [6, 13]:
distances for all the sequences.
based on the distance matrix.
sequences based on the guide tree leaf nodes using dynamic
sequence alignment, construct a profile and align the new sequence
longest branch leaf node is aligned. Hence, MSA is achieved.
1: Progressive Multiple sequence Alignment
DNA sequences is known as the pairwise distance. In this
among the species using Jukes Cantor distance formula.
proposed to measure pairwise distance, the first of
Jukes Cantor distance formula is based on the ratio
gaps in the DNA of two sequences of the species.
D = (-3/4 * log (1-(4/3*R)))
distance between two DNA sequences and `R' is the ratio
non-gap letters [8]. Once we obtain a distance matrix
based on distance matrix is built.
guide tree plays an important role as the branches
be considered for the next step of alignment [10]. A
constructed dependent on the distance matrix of the DNA
evolutionary tree which shows interrelations among various
dependent on physical or genetic characteristic similarities
5,No.3/4, August 2015
2
sequence alignment
pairwise distance
distance matrix
alignment with
progressive alignment
accuracy, namely
Neighbor Join
construction is
programming
sequence to the
achieved.
this project, a
formula. There are a
of which is the
of number of
Jukes cantor
ratio of number
matrix for the
branches of the tree
A guide tree is
DNA sequences.
various biological
similarities and

3
differences. They show the distance between pairs of sequences when the tree edges are
weighted [10]. There are several types of phylogenetic trees; rooted, unrooted, and bifurcating. In
this project we will be using an unrooted phylogenetic tree as our guide tree.Guide trees may be
built using clustering algorithms or other learning models. In this project, the guide tree is
constructed using both UPGMA and Neighbor-Joining algorithms and the final results are
compared. Now we will have a closer look at these two algorithms.
2.2.1. UPGMA
UPGMA stands for Un-weighted Pair-Group Method with Arithmetic mean. Un-weighted refers
to all pairwise distances contributing equally, pair-group refers to groups being combined in pairs,
and arithmetic mean refers to pairwise distances between groups being mean distances between
all members of the two groups considered [7].
Consider four DNA sequences namely: S1, S2, S3 and S4. First, find the pairwise distances
between all the sequences. Then, find the smallest value in the distance matrix and its
corresponding sequences of the shorter distance. For instance let the two sequences with the
shortest distance between them be S1 and S2. Now, cluster S1 and S2 and name the cluster as C1,
updating the distance matrix by eliminating S1 and S2, but including C1. The C1 value
corresponding to the remaining sequences in the distance matrix is calculated with the values of
S1 and S2, i.e., arithmetic mean of S1 and S2 distances with corresponding to the other sequences.
Now, moving forward by considering the updated distance matrix, find the smallest distance
again and its corresponding sequences or clusters (namely sets of sequences). Say this next
smallest distance corresponds to that between clusters C3and C4. Then, in the next step, C3 and C4
would be merged into a new cluster C5, and all the distances to C5 would be updated in the
distance matrix by the corresponding average distances to the sequences in C5. Repeat the same
and find new clusters and sequences, merging and updating the distance matrix, until we are left
with one cluster.
Algorithm: Let the clusters be C1, C2, C3,....., Cn and si be the size of each cluster Ci, ‘d' be the
pairwise distance defined on the clusters. Clustering is done in the following manner:
1. Find the smallest pairwise distance amongst the clusters, d (Ci, Cj).
2. A new cluster Ck with size si+ sj is formed by joining Ci and Cj.
3. Compute the new distances from all other clusters to Ck by using the existing weighted
distances average.
Where l € {1, 2... n} and l ≠ i, j.
4. Repeat 1, 2, and 3 until only one cluster remains.
The asymptotic time complexity of UPGMA is O (n2
), since there are (n-1) iterations, with O (n)
steps per iteration [11].
2.2.2. Neighbor-Joining
The Neighbor-Joining algorithm is consistent with a parsimonious evolutionary model in which
the total sum of branch lengths is minimized [9]. It has the added benefit of achieving the correct
tree topology when given the correct pairwise distances, while also being flexible enough to
accommodate many distance models. Now let us look at the actual algorithm.

Algorithm: Let the clusters be C
pairwise distance defined on the
1. Compute
2. Find the smallest difference
3. A new cluster Ck is formed
between Ci, Cjand Ck as:
4. Compute distances between
Where l € {1, 2...
5. Repeat 1, 2 and 3 until you
Similarly to UPGMA, the asymptotic
O (n2
). However, as we shall see,
2.3. Dynamic Programming
Needleman and Wunsch’s elegant
well for aligning nucleic acid sequences.
algorithms for finding optimal
programming a global alignment
matrix. Paths in the scoring matrix
scoring matrix is dependent on
Match indicates that the two letters
different, and gap (Insertion or Deletion)
Let us consider an example with
score=3, mis-match score= 1, and
optimal solution of aligned sequences.
scoring with match, mis-match, and
Figure 2: Dynamic
C1, C2, C3,....., Cn and si be the size of each cluster
clusters. Clustering is done in the following manner:
difference d (Ci, Cj) – uj – uj and choose i and j accordingly.
formed by joining Ci and Cj. Calculate the intermediate
as:
between new cluster and all the other cluster's:
2... n} and l ≠ i, j.
you are left with two clusters.
asymptotic time complexity of the Neighbor-Joining algorithm
see, their execution time performance is certainly not identical.
elegant algorithm for comparing two protein sequences
sequences. The algorithm actually belongs to a very
optimal solutions called dynamic programming [1].
alignment of two sequences is done based on constructing
matrix decide the optimal solution for two aligned sequences.
three variables; match score, mis-match score, and
letters are the same, mismatch denotes that the two
Deletion) denotes one letter aligning to a gap in the other
with two strings: ATGCG and TGCAT. Consider the scores
and gap penalty= -1. There is no rule that we should have
sequences. Figure 2 shows the complete scoring matrix obtained
and gap scores.
Dynamic Programming pairwise sequence alignment.
5,No.3/4, August 2015
4
Ci, ‘d' be the
manner:
accordingly.
intermediate distances
algorithm is also
identical.
sequences also works
large class of
In dynamic
constructing a scoring
sequences. The
and gap penalty.
two letters are
other string.
scores as match
have only one
obtained after

The best optimal alignment that
2.4. Profiles
A profile is a table that records the
Usually profiles are represented
build a profile we need at least
sequences for two or more sequences
to compute multiple alignment heuristically
used to compute multiple alignments.
detail.
Consider four DNA sequences that
The profile for the above four sequences
Construct a profile sequence: Depending
that position [14] is determined.
chance to get `T'. As the probability
profiled sequence. Moving on the
positions. Then the final profiled
complications in deciding the character,
occurrence? For those situations,
the character at the same position
profile. Then, the rules are as follows:
1. If the character exists at
matched, consider the matched
of the identically highest
Example: Consider that
profile and we have a `
among À', `C’, `T' and
position 4 in the profiled
it will be selected and copied
2. If the character does not
probable character.
that is achieved from above figure are:
A T G C G _
_ T G C A T
the frequency of each letter at each position in a DNA
using a matrix with letters as columns and position
least two DNA sequences. Profiles allows us to identify
sequences that are already aligned. Progressive alignment
heuristically [14]. In this project, the concept of aligned
alignments. Below is an example that explains the concept
that are in a MSA: AGT_C, AGTGC, ATTG_ and TG_GT.
sequences looks like;
Figure 3: Profile Scoring Matrix
Depending on the highest score for each position, the
Suppose we have, at position 0 a 0.75 chance to get
probability of À' is greater than `T', À' is chosen at position
the same way `G', `T', `G’, and `C’ are chosen for
profiled sequence achieved is “AGTGC”. This example
character, but what if two alphabets have same probability
situations, we state and implement two simple rules in this work.
position in the sequence or profile that you want to align to
follows:
at that position, compare it with the highest probable
matched character for the profiled sequence; otherwise,
highest probable characters with a random draw.
À', `C’, `G' and `T' are with 0.25 probability at position
`-' in the sequence at position 4. Then a random draw
and `G'. Suppose that `C' is randomly drawn, then `C'
profiled sequence. If any of À', `C’, `G' and `T' are at that
copied to position 4 of profiled sequence.
not exists at that position, then randomly choose any of
5,No.3/4, August 2015
5
DNA sequence.
position as rows. To
identify consensus
alignment uses profiles
aligned profiles is
concept of profiles in
TG_GT.
he character at
get À' and 0.25
position 0 in the
the next four
example has no
probability of
work. Consider
to the existing
characters. If
otherwise, select any
position 4 in the
draw is made
`C' will be at
position then
of the highest

3. EXPERIMENTAL SETUP
3.1. Experimental Setup
We have discussed how to build
section 2. Now we discuss our
experimental purposes, three types
Input:
1. Seven short sequences
resemble any species but
2. Five sequences with lengths
3. The beta-cassein genetic
and Porpoise.
System Configuration: Intel CORE
Operating System: Ubuntu 14.04
Programming Language: C++
Setup 1:
Match Score: 3
Mis-match Score: 0
Gap penalty: -1
Guide-tree: UPGMA
3.2. Experimental Results
The Figures 4 and 5 are the guide
with input as the large sequences
can clearly observe that the guide
construction algorithms are used.
The progressive multiple alignment
guide tree topology of Figure 5
actual phylogenetic tree of these
Figure 4: UPGMA
Figure 6 shows graphs for execution
sequences as inputs for UPGMA
UPGMA and Neighbor-Joining
short length. This means that the
AND RESULTS
build an effective MSA through the progressive alignment
our implementation and results with some sample data
types of data sets are used:
each with lengths between 4 and 40. These sequences
but are used for testing.
lengths between 40 and 500, also not derived from actual
genetic sequences of five mammalian species Rat, Camel,
CORE i5 with 4GB RAM and 512GB Hard disk space.
14.04
Setup 2:
Match Score: 3
Mis-match Score: 0
Gap penalty: -1
Guide-tree: Neighbor-Joining
guide trees for both UPGMA and Neighbor-Joining,
sequences of rat, camel, dog, whale, and porpoise. From these
guide tree topologies are not same when different evolutionary
used. This has a great impact on the multiple sequence
alignment is purely dependent on the guide tree topology.
for the Neighbor-Joining algorithm is more consistent
species than that of the UPGMA algorithm.
UPGMA Figure 5: Neighbor-Joining
execution time generated by considering short and medium
UPGMA and Neighbor-joining. Observations from the
take the same amount of time when the input sequences
the algorithm used to build the guide tree has no impact
5,No.3/4, August 2015
6
alignment method in
data sets. For
sequences do not
actual species.
Camel, Dog, Whale,
space.
respectively,
these figures, we
evolutionary tree
sequence alignment.
Notably, the
consistent with the
medium length
graph: Both
sequences are of
impact on the

execution time or time complexity
relatively shorter sequences.
Figure
Figure
Figure 7 graphs execution times
and Neighbor-joining. Recall that
camel, porpoise, and whale. The
for both UPGMA and Neighbor-
terms of efficiency in this experiment
Setup 1 with small input sequences:
ACGTACT
ACTACG
ATGGATACTAACTCGG
ATGGCTA_GT
ATGCTCCGGCAAAGG
ATGCTGG
ATCGACAGTGTCA
Thus Multiple Sequence Alignment
Figure 8 below indicates the graph
medium, and large sequences as
graph are as follows: Neighbor-
input sequences. Note that both
actual optimal cost. Therefore,
Joining is more accurate than that
complexity of the program when the progressive alignment
Figure 6: Execution Time for Small sequences
Figure 7: Execution Time for Large sequences
generated by considering large sequences as inputs
that the sample data tested is for genetic sequences
The interesting observation from the graph is that the execution
-Joining differ. Neighbor-Joining clearly outperforms
experiment of a set of long sequences.
sequences: The resulted sequences are:
ACGTAC_T_ _ _ _ _ _ _ _
A_CTAC_G_ _ _ _ _ _ _ _
ATGGATACTAACTCGG
ATGG_ _ _CTA_GT_ _ _
ATGCTCCGGCAAAGG_
ATGCT_ _ _GG_ _ _ _ _ _
ATCGACAGT_ _GTCA_
Alignment is achieved.
graph for total alignment costs generated by considering
inputs for UPGMA and Neighbor-Joining. Observations
-Joining has lower optimal costs than UPGMA for
both algorithms attempt to achieve the lowest total cost,
the results indicate that the final MSA achieved by
that achieved by UPGMA.
5,No.3/4, August 2015
7
alignment is done for
inputs for UPGMA
sequences of rat, dog,
execution time
outperforms UPGMA in
considering small,
Observations from the
for all types of
cost, meaning the
by Neighbor-

Figure 8: Total
4. CONCLUSIONS
Progressive alignment is highly
are arranged by following the leaf
common guide tree algorithms,
accuracy. A simple and greedy
alignment, making the progressive
alignment process previously outlined
profile alignment used in this project
indicate that Neighbor-Joining
alignment, for use in guide trees
ACKNOWLEDGEMENTS
We would like to take this opportunity
McKenney for taking their valuable
REFERENCES
[1] Tatiana A Tatusova and Thomas
nucleotide sequences. FEMS microbiology
[2] Bo Qu and Zhaozhi Wu. An efficient
and Service Science (ICSESS),
2011.
[3] Julie D Thompson, Desmond
progressive multiple sequence
and weight matrix choice. Nucleic
[4] Juan Chen, Yi Pan, Juan Chen,
multiple sequence alignment.
2006. 20th International Conference
[5] Burkhard Morgenstern. Dialign
sequence alignment. Bioinformatics,
[6] Dan Wei and Qingshan Jiang.
construction. In Bio-Inspired
International Conference on, pa
[7] Che-Lun Hung, Chun-Yuan Lin,
three-profile alignment method.
IJCBS’09. International Joint Confer
Total Costs for small, medium, and large sequences
dependent on how the guide tree is built and how the
leaf nodes of a guide tree. In this project, we have compared
UPGMA and Neighbor Joining, in terms of both efficiency
eedy approach is used for profile to sequence and profile
progressive alignments relatively faster. Although the five step
outlined is not new, some aspects of the profile evaluation
project are novel. Furthermore, our experimental results
is superior to UPGMA in both time complexity
for progressive MSA.
opportunity to thank Dr. Xudong William Yu and
valuable time in reviewing the project and providing good
Thomas L Madden. Blast 2 sequences, a new tool for comparing
microbiology letters, 174(2): 247–250, 1999.
efficient way of multiple sequence alignment. In Software
(ICSESS), 2011 IEEE 2nd International Conference on, pages 442
G Higgins, and Toby J Gibson. Clustal w: improving the
sequence alignment through sequence weighting, position-specific
Nucleic acids research, 22(22):4673–4680, 1994.
Chen, We Liu, and Ling Chen. Partitioned optimization
In Advanced Information Networking and Applications,
Conference on, volume 2, pages 5–pp. IEEE, 2006.
Dialign 2: improvement of the segment-to-segment approach
Bioinformatics, 15(3):211–218, 1999.
Jiang. A dna sequence distance measure approach for phylogenetic
Computing: Theories and Applications (BIC-TA), 2010
pages 204–212. IEEE, 2010.
Lin, Yeh-Ching Chung, and Chuan Yi Tang. A parallel
method. In Bioinformatics, Systems Biology and Intelligent Computing,
Conference on, pages 153–159. IEEE, 2009.
5,No.3/4, August 2015
8
the sequences
compared two
efficiency and
profile to profile
step progressive
evaluation and
results clearly
complexity and final
and Dr. Mark
good feedback.
comparing protein and
Software Engineering
442–445. IEEE,
the sensitivity of
specific gap penalties
algorithms for
Applications, 2006. AINA
approach to multiple
phylogenetic tree
2010 IEEE Fifth
parallel algorithm for
Computing, 2009.

9
[8] Conrad Shyu, Luke Sheneman, and James A Foster. Multiple sequence alignment with evolutionary
computation. Genetic Programming and Evolvable Machines, 5 (2):121–144, 2004.
[9] Naruya Saitou and Masatoshi Nei. The neighbor-joining method: a new method for reconstructing
phylogenetic trees. Molecular biology and evolution, 4(4):406–425, 1987.
[10] Li Yujian and Xu Liye. Unweighted multiple group method with arithmetic mean. In Bio-Inspired
Computing: Theories and Applications (BIC-TA), 2010 IEEE Fifth International Conference on,
pages 830–834. IEEE, 2010.
[11] IlanGronau and Shlomo Moran. Optimal implementations of upgma and other common clustering
algorithms. Information Processing Letters, 104(6):205–210, 2007.
[12] Kevin Howe, Alex Bateman, and Richard Durbin. Quicktree: building huge neighbour-joining trees of
protein sequences. Bioinformatics, 18(11):1546–1547, 2002.
[13] Da-Fei Feng and Russell F Doolittle. Progressive sequence alignment as a prerequisitetto correct
phylogenetic trees. Journal of molecular evolution, 25(4):351–360, 1987.
[14] Chikhaoui, Belkacem, Shengrui Wang, and HélenePigot. "Causality-Based Model for User Profile
Construction from Behavior Sequences." Advanced Information Networking and Applications
(AINA), 2013 IEEE 27th International Conference on. IEEE, 2013.
[15] Motahari, Abolfazl, Guy Bresler, and David Tse. "Information theory for DNA sequencing: Part I: A
basic model." Information Theory Proceedings (ISIT), 2012 IEEE International Symposium on. IEEE,
2012.
Authors
Dega Ravi Kumar Yadav born in India, in 1991. Perusing Master’s in Computer
Science at Southern Illinois University, Edwardsville, Illinois, USA. In 2013 published
an International Journal titled “Authenticated Mutual Communication between two
nodes in MANETs”. Areas of interest are Mobile Ad-Hoc networks, Bio-Informatics
and AI.
Dr. GunesErcal, is an Assistant Professor of Computer Science at Southern Illinois
University Edwardsville. Her primary research area is graph theory and its applications
to diverse areas such as network analysis and unsupervised learning. She received her
PhD in Computer Science from the University of California, Los Angeles in 2008. She
received her BS degrees in both Mathematics and Computer Science in 2001 from the
University of Southern California.

A COMPARATIVE ANALYSIS OF PROGRESSIVE MULTIPLE SEQUENCE ALIGNMENT APPROACHES USING UPGMA AND NEIGHBOR JOIN BASED GUIDE TREES

More Related Content

What's hot (18)

Similar to A COMPARATIVE ANALYSIS OF PROGRESSIVE MULTIPLE SEQUENCE ALIGNMENT APPROACHES USING UPGMA AND NEIGHBOR JOIN BASED GUIDE TREES (20)

Recently uploaded (20)

A COMPARATIVE ANALYSIS OF PROGRESSIVE MULTIPLE SEQUENCE ALIGNMENT APPROACHES USING UPGMA AND NEIGHBOR JOIN BASED GUIDE TREES