Multiple Sequence Alignment (MSA)

MSA distinguishes between orthologs (speciation-derived sequences) and paralogs (duplication-derived sequences) to ensure only true evolutionary relatives are aligned.

• Prevents comparison of functionally divergent proteins
• Uses statistical models like profile hidden Markov models (pHMMs) to detect remote homology
• Critical for transferring functional annotations between species
• Example: Aligning human and mouse hemoglobin sequences reveals conserved heme-binding residues

Quantifies the degree of amino acid preservation at each column position, revealing residues under purifying selection pressure.

• Shannon entropy measures positional variability
• Jensen-Shannon divergence compares observed frequencies to background distributions
• Highly conserved positions often indicate catalytic sites, ligand-binding pockets, or structural cores
• Example: Catalytic triad residues (Ser, His, Asp) in serine proteases show near-absolute conservation across all species

Alignment algorithms apply affine gap penalties to model the biological reality that insertions and deletions (indels) occur in contiguous blocks rather than as isolated events.

• Gap opening penalty: High cost for initiating an indel (typically -10 to -12)
• Gap extension penalty: Lower cost for extending an existing gap (typically -1 to -2)
• Prevents biologically meaningless fragmented alignments
• Example: A 5-residue loop insertion is scored as one event, not five independent deletions

Modern MSA tools like Clustal Omega and MAFFT build alignments hierarchically using guide trees derived from pairwise distance matrices.

• Computes all-vs-all pairwise alignments first
• Constructs a neighbor-joining phylogenetic tree
• Aligns sequences progressively from most similar to most divergent
• Reduces computational complexity from O(L^N) to approximately O(N log N)
• Example: MAFFT's FFT-NS-2 algorithm processes 10,000 sequences in minutes on standard hardware

MSA output is converted into a PSSM that captures the amino acid probability distribution at each column, forming the statistical profile of a protein family.

• Each column contains 20 log-odds scores (one per amino acid)
• Used as input features for secondary structure prediction and solvent accessibility models
• Enables sensitive database searching via PSI-BLAST iterations
• Example: A PSSM column with high scores for hydrophobic residues (L, I, V) indicates a buried core position

MSA columns are analyzed for correlated mutation patterns that reveal physically contacting residue pairs in the folded protein structure.

• Direct coupling analysis (DCA) disentangles direct from indirect correlations
• Uses maximum entropy models or sparse inverse covariance estimation
• Predicted contacts serve as distance restraints for ab initio structure prediction
• Example: AlphaFold's predecessor relied heavily on MSA-derived coevolutionary signals to achieve breakthrough contact prediction accuracy

A position-specific scoring matrix is a quantitative profile derived directly from an MSA that captures the frequency or log-odds probability of observing each amino acid at every alignment column. Unlike a simple consensus sequence, a PSSM encodes the full spectrum of evolutionary conservation and variation at each site.

• Construction: Calculated by counting amino acid occurrences per column, often weighted for sequence redundancy and corrected with pseudocounts to avoid zero probabilities.
• Application: Used as input features for predicting secondary structure, solvent accessibility, and protein-protein interaction interfaces.
• PSI-BLAST: The classic iterative search tool uses PSSMs to detect remote homologs by searching sequence databases with a profile rather than a single query sequence.

The BLOcks SUbstitution Matrix is a pre-computed set of log-odds scores for amino acid replacements, derived from conserved ungapped blocks within aligned protein families. The most common variant, BLOSUM62, serves as the default scoring scheme for sequence alignment algorithms.

• Derivation: Scores are calculated from observed substitution frequencies in blocks of sequences sharing no more than a specified percentage identity (e.g., 62% for BLOSUM62).
• Log-Odds: Positive scores indicate substitutions that occur more frequently than expected by chance, reflecting conservative physicochemical replacements.
• Role in MSA: Guides the progressive alignment process by determining the cost of inserting a gap or aligning two divergent residues.

Contact prediction is the task of identifying pairs of amino acid residues that are in spatial proximity (< 8 Å between Cβ atoms) within the folded three-dimensional structure, using evolutionary coupling signals extracted from an MSA.

• Direct Coupling Analysis (DCA): A statistical framework that disentangles direct evolutionary couplings from transitive correlations caused by chains of interacting residues.
• Covariance: The core signal is the correlated mutation pattern—when residue A changes, residue B tends to change to maintain structural integrity.
• Deep Learning Integration: Modern predictors like AlphaFold use raw MSA-derived co-evolutionary features as input to neural networks, dramatically improving 3D structure prediction accuracy.

A profile hidden Markov model is a statistical construct built from an MSA that represents a protein family's sequence consensus, position-specific insertion/deletion probabilities, and amino acid emission frequencies in a unified probabilistic framework.

• Architecture: Consists of match states (consensus columns), insert states (extra residues), and delete states (gaps), with transition probabilities governing movement between them.
• HMMER Suite: The widely used HMMER3 software uses profile HMMs to search sequence databases with significantly higher sensitivity than pairwise methods for detecting remote homologs.
• Pfam Database: Curated profile HMMs define Pfam domains, the gold-standard resource for protein family annotation and domain architecture analysis.

A deep mutational scan is a high-throughput experimental technique that assays the functional impact of thousands of single amino acid substitutions across a protein, generating a comprehensive genotype-phenotype landscape.

• Methodology: A library of variants is created, subjected to a functional selection or screen, and the frequency of each variant is quantified before and after selection via deep sequencing.
• Fitness Scores: Each mutation receives a quantitative fitness score reflecting its effect on function, stability, or binding.
• MSA Connection: DMS data validates predictions from evolutionary conservation in MSAs—positions with low MSA entropy typically show high sensitivity to mutation in DMS experiments.

A sequence logo is a graphical representation of an MSA that displays the conservation pattern at each alignment position using stacked letters, where the total height of each stack reflects the information content and individual letter heights correspond to their observed frequencies.

• Information Content: Measured in bits, calculated as the reduction in uncertainty relative to a background amino acid distribution. A perfectly conserved position has maximum information.
• Visual Encoding: Tall, dominant letters indicate highly conserved residues critical for structure or function; short, mixed stacks indicate variable positions tolerant to mutation.
• WebLogo: The standard tool for generating publication-quality sequence logos, widely used to visualize binding motifs, active sites, and catalytic residues.