- Substitution matrices are fundamental tools in protein sequence alignment and comparative bioinformatics. They assign numerical scores to substitutions between amino acids and help determine whether two protein sequences are evolutionarily or functionally related. During evolution, amino acids may be replaced by other amino acids through mutations in the underlying DNA sequence. Some substitutions occur more frequently than others because amino acids differ in their chemical properties, structural roles, and effects on protein function. Substitution matrices capture these differences and allow alignment algorithms to evaluate conservative and non-conservative amino acid replacements more realistically than a simple match-and-mismatch scoring system.
- A substitution matrix is generally represented as a square table containing scores for every possible pair of amino acids. For the standard 20 amino acids, a conventional matrix contains 20 rows and 20 columns. Each cell represents the score assigned to aligning one amino acid with another. Identical amino acids usually receive positive scores, although the exact value varies according to their background frequencies and the statistical model. Substitutions between chemically similar amino acids may also receive positive scores because they are commonly observed during evolution. Substitutions between chemically dissimilar residues are more likely to receive negative scores. For example, leucine and isoleucine are both hydrophobic amino acids with similar structural properties, so their substitution may receive a favourable score. In contrast, replacing a hydrophobic residue with a strongly charged residue may receive a less favourable score.
- Most commonly used protein substitution matrices are based on log-odds scoring. They compare the probability of observing an amino acid pair in an evolutionary or homologous context with the probability expected from the background frequencies of the amino acids. A general representation is:
[
S(a,b)=\log_2\left(\frac{P(a,b)}{P(a)P(b)}\right)
]
- Here, (S(a,b)) is the substitution score for amino acids (a) and (b), (P(a,b)) represents the probability of observing the pair in the modelled evolutionary context, and (P(a)) and (P(b)) are their background probabilities. The logarithm base and scaling factor may differ between matrix implementations. Positive scores indicate that a residue pair occurs more frequently than expected under the background model, while negative scores indicate that it occurs less frequently than expected. The resulting scores can be used by global and local alignment algorithms to identify biologically meaningful similarities between proteins.
- Two major families of protein substitution matrices are PAM, which stands for Point Accepted Mutation, and BLOSUM, which stands for BLOcks SUbstitution Matrix. Both are widely used in protein sequence alignment, but they were developed using different approaches and datasets. PAM matrices were developed from evolutionary comparisons of closely related proteins and use an evolutionary model to estimate substitutions over increasing evolutionary distances. BLOSUM matrices were developed from conserved regions of protein sequence alignments and use sequence clustering to reduce the influence of closely related sequences. Understanding these differences helps researchers select appropriate matrices for different protein comparison tasks.
- PAM matrices were introduced by Margaret Dayhoff and her colleagues through studies of accepted amino acid substitutions in closely related protein sequences. The original approach examined protein families for which evolutionary relationships could be established with reasonable confidence. An accepted point mutation refers to an amino acid replacement that has persisted in an observed protein lineage, rather than every nucleotide mutation that may have occurred in the underlying DNA. The observed substitutions were used to estimate an amino acid replacement model, which could then be extrapolated to represent greater evolutionary distances.
- PAM1 represents an evolutionary distance corresponding to approximately one accepted point mutation per 100 amino acid positions, under the original model’s definition. It is important to understand that PAM1 does not mean that exactly one percent of positions must differ in every pair of proteins. It is a model-based evolutionary distance. Higher PAM matrices are derived by repeatedly applying the estimated substitution process. For example, PAM250 represents a much greater modelled evolutionary distance than PAM30. The larger the PAM number, the more evolutionary change the matrix is designed to accommodate.
- PAM matrices are particularly useful for understanding evolutionary substitution patterns and comparing proteins across different levels of divergence. PAM30 is commonly associated with relatively closely related proteins and short, highly similar sequence comparisons. PAM70 is useful for somewhat more divergent comparisons, while PAM250 is often used when comparing more distantly related proteins. These are general guidelines rather than strict rules, because the best matrix also depends on sequence length, composition, alignment method, and the biological question.
- The PAM approach is based on an evolutionary substitution probability model. A mutation probability matrix describes the probability that one amino acid will be replaced by another over a specified evolutionary interval. The PAM1 matrix provides the initial substitution model, and higher-distance matrices are obtained through repeated matrix multiplication. In simplified mathematical form:
[
PAM_n = (PAM_1)^n
]
- Here, (PAM_1) represents the one-unit substitution probability matrix, and (n) represents the evolutionary distance. This expression describes the conceptual repeated application of the transition model; actual published PAM scoring matrices also involve background frequencies, log-odds transformations, and score scaling. The resulting matrices estimate how amino acid substitutions accumulate over evolutionary time under the model’s assumptions.
- BLOSUM matrices were developed by Steven and Jorja Henikoff using conserved regions of protein sequence alignments. Their method focused on blocks of conserved amino acid positions found in related proteins. These blocks were selected because they contained regions of meaningful sequence conservation and were suitable for estimating amino acid substitution frequencies. Unlike the PAM approach, BLOSUM matrices were not generated by extrapolating a single close-distance matrix through repeated multiplication. Instead, substitution frequencies were estimated directly from the conserved blocks after sequences were clustered according to their percentage identity.
- The BLOSUM numbering system is based on the sequence identity threshold used for clustering. For example, BLOSUM62 was constructed using a clustering threshold of 62% sequence identity. Sequences exceeding the threshold were grouped to reduce the influence of highly similar sequences when calculating substitution frequencies. The resulting matrix therefore represents substitution patterns estimated from the clustered conserved-block data, not a direct evolutionary distance measured in a fixed number of mutations. BLOSUM45 uses a lower clustering threshold and is generally more suitable for comparisons involving more divergent proteins, whereas BLOSUM80 is commonly used for closely related proteins.
- BLOSUM62 is one of the most widely used general-purpose protein substitution matrices. It is frequently used in BLASTP and many protein alignment applications because it provides a useful balance for a broad range of protein comparisons. BLOSUM80 is often selected for closely related proteins, while BLOSUM45 may be more suitable for more distant relationships. Other matrices, including BLOSUM50 and BLOSUM90, are also available for different comparison requirements. The numbering should not be interpreted as a direct evolutionary distance: a lower BLOSUM number generally corresponds to a matrix intended for more divergent comparisons, but it does not mean that the proteins differ by a particular percentage.
- The main difference between PAM and BLOSUM lies in how their substitution patterns are estimated. PAM matrices originate from observed substitutions in closely related proteins and use an evolutionary model to extrapolate to greater distances. BLOSUM matrices estimate substitution patterns directly from conserved sequence blocks after clustering related sequences at specified identity thresholds. Consequently, PAM numbers increase with modelled evolutionary distance, whereas BLOSUM numbers decrease as the clustering threshold is lowered to accommodate more divergent sequence relationships. Although both matrix families use log-odds scoring, their underlying datasets, estimation methods, and interpretations differ.
- The relationship between PAM and BLOSUM values should not be treated as a simple numerical conversion. For example, PAM250 and BLOSUM45 are both commonly used in comparisons involving relatively divergent proteins, but they are not equivalent matrices. They were developed from different datasets and statistical procedures. The choice between them should be based on the nature of the sequences, the expected evolutionary divergence, the alignment method, and the performance required for the analysis.
- Substitution matrices are particularly important in protein alignment because amino acid replacements are not equally likely. Amino acids can be grouped according to properties such as hydrophobicity, polarity, charge, size, and structural tendencies. Conservative substitutions generally preserve some chemical or structural properties, while radical substitutions produce greater changes. For example, substitutions among leucine, isoleucine, and valine are often more acceptable than substitutions between these hydrophobic residues and strongly charged amino acids. Substitution matrices reflect statistical patterns of such replacements across protein families. However, a favourable substitution score does not guarantee that a replacement is functionally harmless, because the biological effect depends on the specific protein position and its structural or functional environment.
- PAM and BLOSUM matrices are used in both global and local protein sequence alignment. In global alignment, the entire length of two sequences is compared, whereas local alignment identifies the most similar regions within the sequences. Algorithms such as Needleman–Wunsch and Smith–Waterman use substitution scores to evaluate aligned amino acid pairs and combine them with gap-opening and gap-extension penalties. A substitution matrix determines the score of each residue pair, while the gap penalties determine the cost of introducing and extending insertions or deletions. Both components influence the final alignment.
- Substitution matrices are also central to protein similarity searching. BLASTP uses a substitution matrix, together with gap costs and statistical parameters, to identify regions of similarity between a query protein and sequences in a database. BLOSUM62 is a common default matrix for BLASTP, although other matrices can be selected when appropriate. PSI-BLAST uses position-specific scoring information generated from iterative searches to detect more distant protein relationships. Position-specific scoring matrices are related to substitution matrices but are not identical: a substitution matrix provides general scores for amino acid pairs, whereas a position-specific matrix assigns scores according to the residue preferences at individual positions in a particular sequence profile.
- The selection of a substitution matrix should reflect the evolutionary distance between the proteins being compared. For closely related proteins, matrices such as PAM30, PAM70, BLOSUM80, or BLOSUM90 may provide useful sensitivity. For more distant comparisons, PAM250, BLOSUM45, or BLOSUM50 may be appropriate starting points. BLOSUM62 is a widely used general-purpose choice. These recommendations are not absolute. Very short sequences, unusual amino acid composition, low-complexity regions, and highly specialised protein families may require different parameters or additional analysis. Researchers should evaluate alignment coverage, sequence identity, alignment score, statistical significance, and biological context rather than selecting a matrix based only on its name.
- The following Python example demonstrates how to load and inspect PAM and BLOSUM substitution matrices using Biopython. The package provides several standard matrices that can be used in protein sequence alignment.
from Bio.Align import substitution_matrices
# Load substitution matrices
blosum62 = substitution_matrices.load(“BLOSUM62”)
blosum45 = substitution_matrices.load(“BLOSUM45”)
pam250 = substitution_matrices.load(“PAM250”)
# Display selected substitution scores
print(“BLOSUM62: W-W =”, blosum62[“W”, “W”])
print(“BLOSUM62: W-F =”, blosum62[“W”, “F”])
print(“BLOSUM62: W-D =”, blosum62[“W”, “D”])
print(“BLOSUM45: W-F =”, blosum45[“W”, “F”])
print(“PAM250: W-F =”, pam250[“W”, “F”])
- The example retrieves scores for selected amino acid pairs. The values may differ between matrices because each matrix was developed using a different substitution model and dataset. The scores can be inspected to understand how each matrix evaluates identical residues and amino acid replacements.
- The following example demonstrates how to use BLOSUM62 and PAM250 in protein pairwise alignment.
from Bio.Align import PairwiseAligner
from Bio.Align import substitution_matrices
protein1 = “MKTAYIAKQRQISFVKSHFSRQ”
protein2 = “MKTAYIAKQRIISFVKSHFSRQ”
for matrix_name in [“BLOSUM62”, “PAM250”]:
matrix = substitution_matrices.load(matrix_name)
aligner = PairwiseAligner()
aligner.mode = “global”
aligner.substitution_matrix = matrix
aligner.open_gap_score = -10
aligner.extend_gap_score = -0.5
alignments = aligner.align(protein1, protein2)
print(“\nMatrix:”, matrix_name)
print(“Alignment score:”, alignments.score)
print(alignments[0])
- This example compares the same protein sequences using two substitution matrices. The gap penalties are kept constant to make the effect of changing the matrix easier to examine. The alignment scores should not be compared as if they were measurements on a common scale, because PAM and BLOSUM matrices use different scoring systems. The purpose is to demonstrate how matrix selection can influence sequence alignment and scoring.
- Substitution matrices also have limitations. They are statistical models derived from particular datasets and assumptions, so they may not accurately represent every protein family or evolutionary process. A general-purpose matrix may be less suitable for proteins with unusual amino acid composition, membrane proteins, highly conserved active sites, or proteins that have undergone specialised evolutionary changes. Substitution scores also do not fully account for structural context, functional constraints, co-evolution between residues, or the effects of insertions and deletions. In addition, a high alignment score alone does not establish homology or prove that two proteins perform the same function. Statistical significance, sequence coverage, conserved domains, structural evidence, and biological information should be considered together.
- PAM and BLOSUM substitution matrices remain essential resources in modern bioinformatics. They provide a systematic way to evaluate amino acid replacements, improve protein sequence alignment, support database searches, and identify evolutionary relationships among proteins. PAM matrices use an evolutionary model derived from accepted point mutations, while BLOSUM matrices estimate substitutions from conserved protein blocks after sequence clustering. Understanding their construction, numbering systems, scoring principles, and limitations allows researchers to select appropriate matrices and interpret protein alignment results more accurately.