PSSMs

Position-Specific Scoring Matrices (PSSMs) are matrices used in sequence analysis to quantify the likelihood of each residue at each position in a sequence profile. They encode position-specific substitution preferences derived from a multiple sequence alignment of related sequences, allowing a single matrix to capture conservation patterns across the region of interest. PSSMs are widely used to detect distant homologs and to score sequences against a protein family or motif.

Construction typically starts with a curated multiple sequence alignment of related sequences. For each position i, the observed frequency p_i(a) of residue a (amino acid or nucleotide) is estimated. To avoid zero probabilities, pseudo-counts or Bayesian priors are often added. Background frequencies b(a) are used to convert to log-odds scores, commonly expressed as s(i,a) = log2( p_i(a) / b(a) ). The resulting matrix has a row for each position and a column for each possible residue. Gaps can be treated with penalties or separate handling.

Applications include scoring a query sequence by summing the scores for each position, enabling the detection of conserved motifs and remote homologs. PSSMs underpin profile-based searches and motif detection workflows, and are central to iterative methods such as PSI-BLAST, which builds and refines a PSSM from hits in a database to improve sensitivity over successive rounds.

Types and relation to other concepts: PSSMs are commonly constructed for proteins (20 amino acids) but can be created for nucleic acids. They are related to position weight matrices used in motif discovery; both are profile representations that, in many implementations, assume positional independence, an approximation with both strengths and limitations. Limitations include dependence on alignment quality, potential biases from overrepresented sequences, and the independence assumption that may overlook inter-positional correlations.