Bioinformatics: Sequence Analysis and Prediction

Bioinformatics has revolutionized biological research, shifting from single experiments to high-throughput studies and predictive modeling. It involves the analysis of large datasets, always in collaboration with experimental biology. Key tasks in bioinformatics include DNA sequencing, genome annotation, gene and protein function prediction, computational evolutionary biology, and protein structure prediction.

Understanding Sequence Analysis and Prediction in Bioinformatics

At its core, sequence analysis involves comparing biological sequences like proteins and nucleic acids. This comparison is vital for identifying new sequences, understanding functional and structural patterns, assessing similarity, and conducting evolutionary analysis.

Visualizing Sequence Similarity with Dot Plots

Dot plots offer one of the most basic ways to compare sequences. They allow for the quick identification of regions of local alignment, repeats, insertions, and deletions. While useful for visualization, dot plots do not provide a robust statistical measure of similarity.

To create a dot plot, one sequence is placed horizontally and another vertically on a matrix. Dots are placed where characters match. Diagonal lines indicate areas of similarity. This is a word-based method, where word size can be adjusted, or a threshold (e.g., 4/5 characters matching) can be set for placing a dot.

Dot plots can reveal:

• Local alignments: Matches appear as diagonal lines.
• Repeats: In a self-self alignment, repeats appear as extra diagonals, for example, in human mucin which contains exact and slightly different repeats.
• Low complexity regions: These regions, which may include homopolymeric runs or over-representation of certain residues, appear as solid boxes.

Scoring Matrices: Quantifying Sequence Relatedness

Since dot plots lack statistical rigor, scoring matrices are used to provide numerical measures of similarity. These are empirical weighting schemes developed for both nucleotide and amino acid characters.

The Logic Behind Log-Odds Scores

Scoring matrices use log-odds scores ($S_{i,j}$) to represent the ratio of observed versus random frequency of substitution for two residues, i and j. The formula is $S_{i,j} = \log \frac{(q_{i,j})}{(p_i p_j)}$, where $q_{i,j}$ is how often i and j are empirically seen to align, and $p_i$ and $p_j$ are the probabilities of residues i and j occurring randomly.

• Positive scores indicate residues that replace each other more often than by chance. For example, tryptophan (Trp) matching Trp scores high (e.g., 11 in BLOSUM62) due to its rarity and importance.
• Negative scores suggest residues that replace each other less often than by chance.
• Conservative substitutions (e.g., Tyr for Trp, score of 2) are observed more often than by chance but are weaker than exact matches.
• Non-conservative substitutions (e.g., Val for Trp, score of -3) are observed less often than by chance.

Nucleotide Scoring Matrices

The landscape for nucleotide scoring matrices is simpler, often counting matches and mismatches. Equal frequency is usually assumed, though purine-purine versus purine-pyrimidine substitutions may be scored differently. Generally, protein searches are more powerful than nucleotide searches due to the greater complexity and information content of amino acid sequences.

Amino Acid Scoring Matrices: PAM vs. BLOSUM

Amino acid scoring matrices account for biological factors like conservation (absolute conservation and conservative substitutions), frequency of residues, and evolutionary rates. Amino acids tend to have higher replaceability within their physical-chemical groups.

PAM Matrices: Dayhoff's Pioneer Work

PAM (Point Accepted Mutation) matrices, developed by Dayhoff et al. in 1978, are based on substitution patterns in closely related proteins (over 85% similarity) aligned globally. The PAM1 matrix represents approximately 1 amino acid change per 100 residues, or 1% divergence.

Key assumptions and limitations of PAM matrices include:

• Amino acid replacement is independent of previous mutations at the same position.
• Matrices for longer evolutionary distances (e.g., PAM250) are extrapolated predictions, not empirical observations.
• All sites are assumed equally mutable, ignoring conserved motifs.
• Based on a relatively small number of sequences available in 1978, with a bias towards small, globular proteins.
• Assumes evolutionary forces are constant across different timescales.

BLOSUM Matrices: Local Alignment Insights

BLOSUM (Blocks Substitution Matrix) matrices, developed by Henikoff & Henikoff in 1992, are based on conserved areas from the BLOCKS database, using local alignments. These matrices are directly calculated across varying evolutionary distances, not extrapolated.

For example, BLOSUM62 represents sequences with no more than 62% identity, where sequences above this threshold are clustered and their contribution weighted. This approach provides a more empirical basis for evolutionary distances.

Selecting the Right Scoring Matrix

Choosing the appropriate matrix is crucial for accurate sequence analysis. The choice depends on the expected similarity percentage of the sequences being aligned:

Gaps and Gap Penalties: Accounting for Indels

Gaps in sequence alignments represent insertions or deletions (indels), which are evolutionarily expensive events. To avoid biologically implausible scenarios, gaps must be limited. There are two main types of gap penalties: linear and affine.

Affine Gap Penalties: A Parsimonious Approach

Affine gap penalties assume a parsimonious model of evolution, where a single large gap is more likely than multiple small gaps, as it represents a single evolutionary event. The penalty is calculated as G + Ln, where:

• G is the gap opening penalty
• L is the gap extension penalty
• n is the length of the gap

Crucially, G > L (gap opening penalty is greater than gap extension penalty). This approach encourages fewer, larger gaps rather than many small ones, reflecting the simpler evolutionary explanation.

FAQ: Bioinformatics Sequence Analysis for Students

What is bioinformatics and why is sequence analysis important?

Bioinformatics is an interdisciplinary field that develops methods and software tools for understanding biological data. Sequence analysis, a core component, is crucial for identifying new genes or proteins, understanding their function and structure, comparing species, and tracing evolutionary relationships. It helps researchers make sense of the vast amount of genetic and protein data generated by modern high-throughput studies.

What are scoring matrices used for in sequence alignment?

Scoring matrices assign numerical values to reflect the likelihood of observing a match or a mismatch between two residues in an alignment. They are essential for quantifying the similarity between sequences and distinguishing meaningful biological relationships from random chance. Different matrices are used depending on the evolutionary distance between the sequences being compared.

What is the difference between PAM and BLOSUM matrices?

PAM matrices (Point Accepted Mutation) are based on global alignments of closely related sequences and are extrapolated to predict scores for more divergent sequences. BLOSUM matrices (Blocks Substitution Matrix) are derived directly from conserved blocks of local alignments across various evolutionary distances. Generally, BLOSUM matrices are considered more empirical and are often preferred for their ability to handle diverse sequence relationships.

Why are gap penalties necessary in sequence alignment?

Gap penalties are crucial because they account for insertions or deletions (indels) that occur during evolution. Without penalties, alignment algorithms might introduce many biologically unrealistic gaps to maximize matches. Penalties, especially affine gap penalties, ensure that alignments reflect a more parsimonious evolutionary history, favoring fewer and larger indels, which are more likely to represent single evolutionary events.

When should I use a dot plot versus a scoring matrix for sequence comparison?

You should use a dot plot for a quick, visual inspection of sequence similarity, to identify regions of local alignment, repeats, or rearrangements, and to get an initial sense of sequence relationships. However, for a statistically robust and quantifiable measure of sequence similarity, especially for evolutionary analysis or database searching, scoring matrices with appropriate alignment algorithms (like BLAST or FASTA) are necessary.