Scoring Schemes

Learning Objective
This tutorial introduces you to the scoring systems that can be used in SeqAn to quantify the sequence similarity. You will learn basic techniques to create and modify standard and custom scoring systems capable to satisfy the requirements of a wide range of applications.
Difficulty
Basic
Duration
45 min
Prerequisites
A First Example, Sequences, Alignment Representation (Gaps)

The alignment procedures are usually based on the sequences similarity computation described by an alignment scoring system that gives countable information used to determine which sequences are related and which are not.

Four main biological events must be considered during the sequence alignment: Conservation, substitution, insertion and deletion. We could have a Conservation when the two compared letters are the same and a Match is detected, a Substitution when we detect a Mismatch where a letter is aligned with another, and Insertion or Deletion when in one of the two aligned sequences a letter is aligned with a Gap. Matches, mismatches and gaps detected during the alignment do not guarantee to be the most representative biological truth since their dispositions is dependent of the chosen scoring schemes and the selected alignment algorithm. In order to improve the correlation between computed sequence alignment and biological similarity, specific combinations of scoring schemes and alignment algorithms have been developed during the years and are usually adopted for the alignment of different types of biological sequences. For example, as we will see in the following, the small RNA sequences are usually aligned with a Global Alignment algorithm implementing a Simple Score scheme, differently from the protein sequences that are mostly aligned with the Local Alignment algorithm that uses a Substitution Matrix Score scheme.

../../../_images/alig_ins_del.png

Given an alignment structure that store the two sequences and a scoring scheme, the score of the alignment can be computed as the sum of the scores for aligned character pairs plus the sum of the scores for all gaps.

With refer to the alignment procedure a Scoring Scheme can be defined as the set of rules used to assess the possible biological events that must be considered during the alignment procedure.

In SeqAn are available several scoring schemes to evaluate matches and mismatches, while three different gap models can be applied to consider insertions and deletions events. We will first introduce you to the scoring schemes used to evaluate match and mismatch. Subsequently, you will learn how to chose the gap model to be implemented in the chosen scoring scheme.

Match/Mismatch Evaluation

Simple Score

The simplest example of Scoring Scheme, usually applied to score the similarity among nucleotide sequences, is the Levenshtein distance model that assigns a score of 0 and -1 respectively if a match or a mismatch occurs, whereas a penalty value equal to -1 in case of gaps representing insertions or deletions (this scoring scheme is the default for SimpleScore). Alternatively, also the Hamming distance model can be used for some simple tasks that do not require the gap evaluations.

Now, let’s start by constructing our first scoring function for the global alignment algorithm called with the function globalAlignment. As first step we need to include the header file <seqan/align.h> which contains the necessary data structures and functions associated with the alignments. The next steps would be to implement the main function of our program and to define the types that we want to use.

#include <iostream>
#include <seqan/align.h>

using namespace seqan;

int main()
{
    typedef String<AminoAcid> TSequence;             // sequence type
    typedef Align<TSequence, ArrayGaps> TAlign;      // align type

We first define the type of the input sequences (TSequence) and an Align object (TAlign) type to store the alignment. For more information on the Align datastructure, please read the tutorial Alignment Representation (Gaps). After defining the types, we can continue to construct our own Align object. First, we create two input sequences seq1 = "TELKDD" and seq2 = "LKTEL", then we define the scoring values for match, mismatch, gap. As last we create the ‘align’ object and resize it to manage two Gaps objects, at this point we filled it with the sequences to be aligned.

    TSequence seq1 = "TELKDD";
    TSequence seq2 = "LKTEL";
    int match = -0;
    int mismatch = -1;
    int gap = -1;

    TAlign align;
    resize(rows(align), 2);
    assignSource(row(align, 0), seq1);
    assignSource(row(align, 1), seq2);

Now, we can compute the global alignment that makes use of the simple scoring function. To do so, we simply call the function globalAlignment and give as input parameters the align object and the scoring scheme representing the Levenshtein distance. The globalAlignment function fills the align object with the best computed alignment and returns the maximum score which we store in the score variable. Afterwards, we print the computed score and the corresponding alignment.

    int score = globalAlignment(align, Score<int, Simple>(match, mismatch, gap));
    std::cout << "Score: " << score << std::endl;
    std::cout << align << std::endl;

    return 0;
}

Congratulations! You have created your global alignment implementing the simple scoring function, the output is as follows:

Score: -5
      0     .   
        TELK-DD
          ||   
        --LKTEL

However, in the evaluation of protein similarity or for advanced nucleotide alignments a more complex scoring model is generally applied. It is based on the usage of a Substitution Matrix, proven to better describe from a biological point of view, events such as matches and mismatches.

Substitutional Matrices Score

Substitutional Matrices are built on the basis of the probability that a particular amino acid or nucleotide is replaced with another during the evolution process. They assign to each pair a value that indicates their degree of similarities, obtained thanks to statistical methods reflecting the frequency of a particular substitution in homologous protein or RNA families. A positive value in the Substitutional Matrix means that the two letters share identical or similar properties.

These scoring schemes store a score value for each pair of characters. This value can be accessed using score. Examples for this kind of scoring scheme are Pam120 and Blosum62. Anyway the class MatrixScore can be used to store arbitrary scoring matrices for the creation of custom scoring systems, as shown in the example proposed in the Working With Custom Score Matrices.

Blosum matrix, is one of the most used Substitutional Matrix implemented by considering multiple alignments of evolutionarily divergent proteins, while Ribosum is the RNA counterpart computed using ribosomal sequences.

In the following example it is proposed the construction of a scoring function for a global alignment algorithm that uses the Blosum62 matrix to score the matched and mismatched letters. As first we include the header file <seqan/align.h> which contains the necessary data structures and functions associated with the alignments, then we implement the main function of our program and define the types that we want to use.

#include <iostream>
#include <seqan/align.h>

using namespace seqan;

int main()
{
    typedef String<AminoAcid> TSequence;             // sequence type
    typedef Align<TSequence, ArrayGaps> TAlign;      // align type

The input sequences type TSequence and the Align object of type TAlign are defined and the two input sequences seq1 = "TELKDD" and seq2 = "LKTEL" together with the gap penalty are assigned. In this case we define only the gap value since the Blosum matrix will be used to score matches and mismatches. Then the sequences are associated with the alignment object.

    TSequence seq1 = "TELKDD";
    TSequence seq2 = "LKTEL";
    int gap = -1;

    TAlign align;
    resize(rows(align), 2);
    assignSource(row(align, 0), seq1);
    assignSource(row(align, 1), seq2);

Now, we compute the global alignment function, providing as second parameter the tag referred to the Blosum62 matrix together with the gap costs. To do so, we simply call the function globalAlignment and give as input parameters the align object and the Blosum62 scoring scheme. The globalAlignment function returns the score of the best alignment, which we store in the score variable that is then printed together with the corresponding alignment.

    int score = globalAlignment(align, seqan::Blosum62(gap, gap));
    std::cout << "Score: " << score << std::endl;
    std::cout << align << std::endl;

    return 0;
}

The output of a global alignment implementing the Blosum62 scoring function is as follows:

Score: 9
      0     .    
        --TELKDD
          |||   
        LKTEL---

Note

As can be noted the output of this scoring scheme is completely different with respect to the output generated with the simple scoring scheme confirming that the scoring scheme choice is one of the most important step to achieve high quality alignments.

Gap Evaluation

In the previous sections we proposed two simple code examples useful to highlight the differences between two scoring schemes capable to evaluate match and mismatch events. In this section we will see the three gap models, implemented in the SeqAn library, to evaluate the insertion and deletion events.

Linear Gap Model

The easiest is the Linear gap model that considers, for the alignment score computation, the gap length (g) giving the possibility to evaluate with different scores gaps of different sizes;

../../../_images/linear.png

This gap model is chosen as standard when only a gap value is provided in the scoring function or when the two provided gaps have the same value. For instance, this gap model as been adopted during the alignment computation of the two proposed examples.

Affine Gap Model

It has been proven that the first amino acid or nucleotide inserted/deleted (identified as gap open) found during the alignment operations is more significant, from a biological point of view, than the subsequent ones (called gap extension), making the so called Affine Gap model a viable solution for the alignment of biomolecules [Car06]. Affine gap model that attribute different costs to the gap open (d) and the gap extension (e) events, is able to assign an higher penalty to the gap presence with respect to its relative length (g).

../../../_images/affine.png

The Affine Gap model implemented in the DP alignment algorithms is however quite expensive both in terms of computational time as well as in terms of memory requirements with respect to other less demanding solutions such as the Linear Gap model application.

Dynamic Gap Model

In SeqAn is provided an optimised version of the Affine Gap model called Dynamic Gap Selector (DGS) designed by Urgese et al. [UPA+14]. This new gap model can be used to reduce the computational time and the memory requirement while keeping the alignment scores close to those computed with the Affine Gap model. The usage of Dynamic Gap model in the Global alignment computation of long strings can give results slightly different from those computed using Affine Gap model since the alignment matrix became bigger and different alignment paths can be chosen during the alignment procedure. Score variation are rare when Dynamic Gap model is used in the Local alignments.

Example Affine vs Dynamic

Tip

The order of the different costs in the scoring scheme is match, mismatch, gapExtend and gapOpen. The gap model selection can be done providing one of the three specific tags (LinearGaps(), AffineGaps() or DynamicGaps()) as last parameter in the scoring function creation. If you want to use Linear Gap costs you could also omit the last parameter gapOpen and the scoring scheme would automatically choose the Linear Gap cost function. The Affine Gap model is chosen as standard when the gap costs are different and the gap model tag is not provided. If the Dynamic Gap model is required the relative tag must be supplied.

In the following we propose an example where two different scoring functions have been created to show how to call a global alignment algorithm that uses the Blosum62 plus the AffineGaps() and DynamicGaps() specializations. The inclusion of the header and the type definition is identical to the previous examples.

#include <iostream>
#include <seqan/align.h>

using namespace seqan;

int main()
{
    typedef String<AminoAcid> TSequence;             // sequence type
    typedef Align<TSequence, ArrayGaps> TAlign;      // align type

The input sequences type and the Align object of type TAlign are then create and initialized. As can be noted we define two different gap values, one for the gap extension and one for the gap open. Even in this example the Blosum62 will be used to score match and substitutions events.

    TSequence seq1 = "TELKDD";
    TSequence seq2 = "LKTEL";
    int gapExtend = -2;
    int gapOpen = -10;


    TAlign alignAffine;
    resize(rows(alignAffine), 2);
    assignSource(row(alignAffine, 0), seq1);
    assignSource(row(alignAffine, 1), seq2);

    TAlign alignDynamic;
    resize(rows(alignDynamic), 2);
    assignSource(row(alignDynamic, 0), seq1);
    assignSource(row(alignDynamic, 1), seq2);

Now, we can compute the global alignment function providing as second parameter the tag referred to the Blosum62 matrix filled with the two different gap costs. Moreover, the tag for the gap model selection is provided. To do so, we simply call the function globalAlignment and give as input parameters the align object, the Blosum62 scoring scheme and the AffineGaps() or DynamicGaps() tag. The globalAlignment function output is then printed.

    int scoreAffine = globalAlignment(alignAffine, Blosum62(gapExtend, gapOpen), AffineGaps());
    std::cout << "ScoreAffine: " << scoreAffine << std::endl;
    std::cout << alignAffine << std::endl;

    int scoreDynamic = globalAlignment(alignDynamic, Blosum62(gapExtend, gapOpen), DynamicGaps());
    std::cout << "ScoreDynamic: " << scoreDynamic << std::endl;
    std::cout << alignDynamic << std::endl;

    return 0;
}

The output of a global alignment implementing the Blosum62 with the two gap models is as follows:

ScoreAffine: -12
      0     .   
        TELKDD-
          ||   
        --LKTEL



ScoreDynamic: -12
      0     .   
        TELKDD-
          ||   
        --LKTEL

Tip

The functions scoreMatch and scoreMismatch access values for match and mismatch. The function scoreGap, or scoreGapExtend and scoreGapOpen access values for gaps.