Protein Sequence Analysis and Metagenomics

Protein Sequence Analysis

• Data is exchanged between databases nightly, maintaining consistent core data.
• Initial DNA data is translated into protein sequences (e.g., GenBank to GenPept, EMBL to TrEMBL).
• Swissprot is a high-quality annotation source for some sequences, part of UniProtKB.
• UniProtKB (from Swissprot) has a vast database expansion, currently 240 million entries.

Database Issues

• Database organization changes rapidly.
• Protein naming is variable and error correction is slow.
• Database submissions sometimes aren't corrected without submitter action.
• Database access may be browser-dependent (problematic).

Metagenomics

• Craig Ventor pioneered large-scale microbial sequence acquisition in diverse environments (like ocean or gut).
• Metagenomics aims to study biodiversity.
• Data quality and often fragmentary sequences are common in metagenomic studies.
• MGnify (at the EBI) holds 350,000 amplicons from 33,000 metagenomes.

Evolutionary Relationships

• Evaluating species similarity is necessary for detecting evolutionary relationships.
• Pairwise protein sequence alignment is used to quantify similarity.
• Gene duplication and speciation are two vital evolutionary processes causing changes.

Gene Duplication

• Duplication occurs within a genome, with duplicate copies becoming paralogs.
• Duplicate genes result in a possible change of function while one copy maintains the original function.
• These changes in function/structure allow new functions for the duplicated gene.

Speciation

• Speciation results in new species, with each species having a single copy of the same gene, leading to orthologues.
• Orthologues are the resulting protein sequences from a speciation event, with single copies that represent the original gene
• These orthologues can have similar functions, showcasing the common ancestry of the new species.

Pairwise Protein Sequence Alignment

• A scoring scheme and algorithm are required to quantify similarity between amino acid residues in sequences.
• The aim is to create the best alignment possible that captures relevant biological information.

Scoring Scheme

• Scoring 1 point for identical amino acids and 0 for different (identical nucleotides or bases can also be scored this way).
• Protein evolution constrains allowed amino acid changes, leading to largely conservative substitutions (similar chemical properties).
• This maintains the protein structure and function during evolution.
• Chemical property maintenance is known as conservative substitution (e.g., non-polar residues tend to stay non-polar).
• Point Accepted Mutation (PAM) matrix, from Dayhoff, models residue changes in closely related sequences.
• (Extends to more distant sequences by assuming matrices can be multiplied.)
• Model PAM250 with 20% identity.

PAM 250 Matrix

• Quantifies the odds of one residue's mutation to another residue.
• "Odds equals observed probability of amino acid residue exchanging over probability of the exchange occurring by chance"
• Odds represents the likelihood of a residue resisting mutation.

BLOSUM62 Matrix

• Derived from aligned protein families (BLOCKS) with 62% pairwise identity, focusing on conserved regions.
• This creates the most accurate and sensitive matrix.
• Used in BLAST/PSI-BLAST family of database searching algorithms.
• Identifies favourable residues and substitutions to find related sequences.
• More accurate to find distantly related sequences due to excluding unaligned loops regions of protein structure.

Gap Penalties

• Gaps or indels (insertions or deletions are penalized, since they reflect evolutionary changes.)
• Gap open penalty (o) is larger than the gap extension penalty (e).

Alignment Methods

• Proteins often consist of domains that are evolutionary units.
• Proteins are composed of structurally distinct parts/domains.

Global vs Local Alignment

• Global alignment aims to align entire sequences (suitable for sequences assumed to be homologous).
• Local alignment finds similar regions within sequences (more suitable for distantly related sequences).

Needlman-Wunsch Algorithm

• General algorithm for sequence comparison to find maximum matches.
• Assesses maximal possible matches, accounting for gaps.
• Iterative matrix method uses a 2D array that represents every residue possibility.
• Three main steps: Assign values, find best pathways from initial values in the matrix back to the beginning, create an alignment based from the highest score.

Database Searching

• A query sequence and database of sequences are required to find similar sequences.
• PSI-BLAST is a widely used program for advanced sequence analysis and searches.
• BLAST searches rapidly find short segments matching databases, but doesn’t guarantee optimal alignments.
• Databases have multiple features like sequences (high level matches) and specific regions with specific functions, etc.

BLAST

• A tool that finds high-scoring pairs (HSPs) from the query sequence to database's short segments.
• Uses BLAST algorithm for fast local search.
• Uses precision and recall to identify better matches.
• Measures match quality for a given sequence.
• Uses statistical significance for true match identification.

Accuracy of Database Searches

• P-value: Probability of getting a given score by chance (smaller is better).
• E-value: Expected number of matches with equivalent or higher scores by random chance.

MMSeq2

• Designed to handle the increase speeds in DNA sequencing throughput.
• Designed for large-scale metagenomic and related DNA studies.
• Improves speed-sensitivity for searches and provides improved metagenomic analysis.
• Three stage process: k-mer matches, ungapped alignment, full gapped alignment, which accelerates the matching speed while allowing for more accurate results from remote matches.

Recall and Precision

• Recall (or Sensitivity):Fraction of actual positives that are identified.
• Precision: What fraction of positives called as positive are actually positive.
• Specificity: Fraction of negatives that are correctly identified as negatives.

ROC Curves

• Graphical representation showing different thresholds/cut-offs for sensitivity and specificity and useful for comparative analysis of algorithms.

Protein Families

• Related proteins with similar sequences, structures, and functions.
• Useful for studying evolutionary relationships and structural/functional analysis.
• Key residues tend to be conserved to maintain function.
• Multiple sequence alignment for studying protein families is useful.

CLUSTAL

• A heuristic algorithm (educated guess) to create multiple alignments.
• Iterative approach starting with nearest related sequences.

Other Programs

• Clustal Omega (more advanced).
• MUSCLE (fast algorithm using short sequence region words for approximate matches)
• T-Coffee.

PROSITE

• Database of protein motifs.
• Useful for searching for specific sequence patterns with motifs.

Sequence Profiles and Hidden Markov Models (HMMs)

• Sequence profiles (PSSMs): Position-specific scoring matrices.
• HMMs: Probabilistic models using scoring matrices (PSSM's).
• More sensitive for large sequence regions than PROSITE.

PSI-BLAST

• Advanced BLAST variant using existing alignments to improve searches for distant relationships.
• Iterative procedure for finding better matches, improving accuracy.
• Useful tool for identifying/testing homology relationships.
• Creates new profiles and adds to previously formed profiles.

Next-Generation Sequence Analysis

• More sensitive methods to find distant homologues: HH-BLAST, JACKHMMER