MSA Parsing - Usage Guide

Overview

This skill focuses on parsing and analyzing multiple sequence alignments (MSAs). It covers extracting information, analyzing gaps and conservation, filtering sequences, and preparing alignments for downstream analysis like phylogenetics or structure prediction.

Prerequisites

pip install biopython numpy

# Optional for Pfam-scale streaming:
pip install pyhmmer

Quick Start

Tell your AI agent what you want to do: - "Find all the fully conserved positions in this alignment" - "Remove sequences with more than 10% gaps" - "Generate a consensus sequence from this alignment"

Example Prompts

Analyzing Content

"Show me the composition of each column in the alignment"

"Find positions that are conserved in at least 80% of sequences"

"Count the gaps in each sequence"

Filtering and Cleaning

"Remove columns with more than 50% gaps"

"Filter out sequences with too many gaps"

"Remove duplicate sequences from the alignment"

Extracting Information

"Get the sequence for species_A from this alignment"

"Extract columns 100-200 from the alignment"

"List all sequence IDs in the alignment"

Consensus and Conservation

"Generate a consensus sequence with 70% threshold"

"Find the most conserved regions in this alignment"

"What is the consensus at each position?"

What the Agent Will Do

Load the alignment file
Parse alignment structure (sequences, columns, gaps)
Perform requested analysis or filtering
Return results (statistics, filtered alignment, consensus)
Optionally save modified alignment

Key Concepts

Term	Description
Column	Vertical slice (same position across all sequences)
Conservation	Fraction of sequences with same residue at a position
Consensus	Most common character at each position
Gap	Missing data represented by '-'

Working with Annotations

Stockholm-derived alignments expose secondary-structure markup, GC/GR per-column annotations, and per-sequence metadata via record.annotations, record.letter_annotations, and alignment.column_annotations:

from Bio import AlignIO

alignment = AlignIO.read('pfam.sto', 'stockholm')
for record in alignment:
    if 'secondary_structure' in record.letter_annotations:
        print(record.id, record.letter_annotations['secondary_structure'])

ss_cons = alignment.column_annotations.get('secondary_structure')

GC SS_cons (consensus secondary structure), GC RF (reference coordinates), and GS metadata (organism, taxonomy) survive read/write through the 'stockholm' format string but are silently discarded when writing to FASTA, PHYLIP, or NEXUS. Keep a Stockholm master copy if annotations matter for downstream analysis.

Tips

Always check gap content before phylogenetic analysis. Columns with >50% gaps often indicate alignment artifacts or inclusion of non-homologous sequences
Gappy columns can reflect guide tree topology rather than true evolutionary events; divergent sequences disproportionately introduce gaps
For trimming before phylogenetics, prefer ClipKIT (kpic-smart-gap mode) over traditional removal tools; aggressive trimming (>20-30% of sites) can hurt tree quality
Conservation thresholds depend on alignment diversity. 80% conservation in a 5-sequence alignment means less than 80% in a 500-sequence alignment
For critical analyses, quantify alignment reliability per column using GUIDANCE2 or MUSCLE5 ensemble before phylogenetic inference
How gaps are handled downstream matters: missing data (default) can be statistically inconsistent; consider indel coding for maximum phylogenetic signal
Stockholm format preserves annotations other formats lose
Position numbering is 0-based (first column is 0)