BLOSUM Substitution Matrix

BLOSUM matrices use log-odds ratios to score amino acid substitutions. Each cell contains the logarithm of the ratio between the observed frequency of a substitution in conserved blocks and the expected frequency by chance.

• Positive scores: Substitutions occurring more often than random chance (e.g., leucine → isoleucine)
• Zero scores: Substitutions at random expectation
• Negative scores: Substitutions occurring less often than chance (e.g., tryptophan → proline)

This statistical framework ensures that alignments reflect true evolutionary relationships rather than spurious similarities.

The number in a BLOSUM matrix name indicates the sequence identity threshold used during construction. Sequences above this threshold are clustered and treated as a single sequence to reduce overrepresentation.

• BLOSUM62: Clusters sequences at ≥62% identity. The default matrix for BLAST and most alignment tools, optimized for detecting distant homologs
• BLOSUM80: Clusters at ≥80% identity. Better suited for closely related sequences with fewer evolutionary divergences
• BLOSUM45: Clusters at ≥45% identity. Designed for highly divergent sequences where only remote homology remains detectable

Lower thresholds tolerate more divergence; higher thresholds favor conservation.

BLOSUM matrices are built from the BLOCKS database, which contains ungapped, multiply-aligned segments corresponding to the most highly conserved regions of protein families.

Construction process:

• Extract all conserved blocks from aligned protein families
• Count all observed amino acid pair frequencies across aligned columns
• Cluster sequences above the chosen identity threshold to avoid bias
• Calculate target frequencies and background probabilities
• Compute log-odds scores: S(a,b) = 2 × log₂(q(a,b) / p(a)p(b))

This data-driven approach ensures the matrix reflects empirically observed substitution patterns rather than theoretical models.

BLOSUM matrices encode biochemical conservation directly into their scoring structure. Amino acids with similar physicochemical properties receive higher substitution scores.

Key groupings reflected in scores:

• Hydrophobic cluster: Leucine, Isoleucine, Valine, Methionine share high mutual scores
• Aromatic cluster: Phenylalanine, Tyrosine, Tryptophan show elevated substitution rates
• Positive charge cluster: Lysine, Arginine, Histidine exhibit moderate interchangeability
• Small/polar cluster: Serine, Threonine, Asparagine, Glutamine
• Cysteine: Scores conservatively due to its unique disulfide bond role

This biochemical encoding makes BLOSUM matrices superior to simple identity-based scoring for detecting functional homologs.

BLOSUM matrices work in concert with gap penalties to produce biologically meaningful alignments. The matrix scores substitutions while gap penalties control insertions and deletions (indels).

Critical relationship:

• Gap penalties must be calibrated to the specific BLOSUM matrix in use
• BLOSUM62 typically pairs with a gap existence penalty of -11 and extension penalty of -1
• Higher-numbered matrices (BLOSUM80) often use more permissive gap penalties since sequences are already closely related
• Mismatched matrix and gap parameters produce suboptimal alignments

The affine gap model (separate existence and extension costs) is standard, penalizing gap initiation more heavily than gap lengthening to favor fewer, longer indels.

BLOSUM and PAM (Point Accepted Mutation) matrices represent two distinct philosophies in substitution matrix design.

BLOSUM advantages:

• Derived from conserved blocks rather than global alignments of closely related sequences
• Directly uses observed frequencies without extrapolation
• BLOSUM62 outperforms PAM250 for most database search tasks
• Better captures local conservation patterns

PAM characteristics:

• Based on an explicit evolutionary model with extrapolation to different distances
• PAM1 represents 1% accepted mutations; higher numbers extrapolate further
• Historically important but largely superseded by BLOSUM in modern pipelines

Modern tools default to BLOSUM62, though PAM matrices retain niche applications in phylogenetic reconstruction.

The foundational input for BLOSUM construction. An MSA is a computational alignment of three or more evolutionarily related protein sequences used to identify conserved regions. BLOSUM matrices are derived directly from the ungapped, highly conserved blocks within these alignments, ensuring that the calculated substitution frequencies reflect strong structural and functional constraints rather than random drift.

The mathematical core of the BLOSUM matrix. Each cell value is a log-odds ratio comparing two probabilities:

• The observed frequency of a specific amino acid pair substituting for one another in conserved blocks.
• The expected frequency of that pair occurring by random chance based on the background amino acid distribution. A positive score indicates a substitution favored by evolution; a negative score indicates a disfavored substitution.

A dynamic evolutionary profile representing a specific protein family, often generated using BLOSUM as a prior. While BLOSUM provides a global average model of substitution, a PSSM captures position-specific conservation patterns. It records the frequency or probability of each amino acid at every column in an MSA, enabling highly sensitive database searches like PSI-BLAST.

The defining parameter of a BLOSUM matrix, indicated by its number (e.g., BLOSUM62). Sequences in the input MSA are clustered if they share more than a threshold percentage of identical residues.

• BLOSUM62: Clusters sequences at 62% identity, making it the standard for general database searches.
• BLOSUM80: Clusters at 80% identity, optimized for finding closely related sequences.
• BLOSUM45: Clusters at 45% identity, designed to detect highly divergent, distant homologs.

Two distinct philosophies for modeling protein evolution:

• PAM (Point Accepted Mutation): An extrapolative model based on closely related sequences, projecting evolutionary change over time using a Markov chain.
• BLOSUM (BLOcks SUbstitution Matrix): An interpolative model derived directly from observed substitutions in conserved blocks of divergent sequences. BLOSUM matrices have largely superseded PAM in modern bioinformatics due to their direct empirical grounding in functional protein regions.

The primary application of BLOSUM matrices in algorithms like Smith-Waterman (local) and Needleman-Wunsch (global). The matrix provides the substitution score for aligning two amino acids, while gap penalties (affine or linear) subtract from the score for insertions or deletions. The optimal alignment is the path that maximizes the total sum of substitution scores and gap penalties, quantifying biological similarity.