Tutorial 11: Reduced Amino Acid Alphabets

This tutorial demonstrates how to convert amino acid sequences to reduced alphabets based on physicochemical properties. Reduced alphabets simplify protein sequences by grouping similar amino acids together, which can improve computational efficiency and reduce noise in machine learning applications.

Learning Objectives

By the end of this tutorial, you will understand:

What reduced amino acid alphabets are and why they are useful
How to convert protein sequences to different reduction schemes
The differences between various reduction strategies
Applications of reduced alphabets in sequence analysis

Background

The standard genetic code uses 20 amino acids, each with distinct physicochemical properties. However, for many computational analyses, this diversity can introduce noise and computational complexity. Reduced amino acid alphabets group similar amino acids based on properties like:

Hydrophobicity: How water-repelling or water-attracting an amino acid is
Charge: Positive, negative, or neutral electrical charge
Size: Physical size of the amino acid side chain
Aromaticity: Presence of aromatic rings in the structure
Structure: Special properties like flexibility (Gly) or rigidity (Pro)

Research has shown that reduced alphabets can:

Improve protein fold recognition
Reduce noise in machine learning models
Speed up sequence comparisons
Simplify pattern discovery in protein sequences

Setup

import Pkg
if isinteractive()
    Pkg.activate("..")
end

import Test
import Mycelia
import BioSequences
import Statistics

Part 1: Basic Conversion

Let's start with a simple example using a fragment of human insulin A chain.

# Example protein sequence (fragment of human insulin A chain)
insulin_sequence = BioSequences.LongAA("GIVEQCCTSICSLYQLENYCN")
println("Original sequence: ", insulin_sequence)
println("Length: ", length(insulin_sequence))
println()

Part 2: Exploring Available Schemes

Mycelia provides six well-established reduction schemes from the literature. Let's list them all:

println("Available reduction schemes:")
for scheme in Mycelia.list_reduced_alphabets()
    info = Mycelia.get_reduced_alphabet_info(scheme)
    println("  $scheme ($(info[:classes]) classes): $(info[:name])")
end
println()

Part 3: Binary Hydrophobic-Polar (HP2)

The simplest reduction divides amino acids into just two classes:

H (Hydrophobic): Water-repelling amino acids (A, C, F, I, L, M, V, W)
P (Polar): Water-attracting amino acids (G, T, S, Y, P, N, D, E, Q, K, R, H)

This is the most aggressive reduction and is useful for identifying hydrophobic core regions versus surface-exposed regions in proteins.

println("=" ^ 70)
println("HP2: Binary Hydrophobic-Polar Reduction")
println("=" ^ 70)
reduced_hp2 = Mycelia.reduce_amino_acid_alphabet(insulin_sequence, :HP2)
info = Mycelia.get_reduced_alphabet_info(:HP2)
println("Description: $(info[:description])")
println("Groups:")
for (letter, aas) in sort(collect(info[:groups]))
    println("  $letter: $aas")
end
println("\nOriginal: $insulin_sequence")
println("Reduced:  $reduced_hp2")
println()

Part 4: Three-Class Hydropathy (HYDROPATHY3)

A more nuanced approach uses three classes based on IMGT hydropathy:

H (Hydrophobic): Strongly water-repelling
N (Neutral): Neither strongly hydrophobic nor hydrophilic
P (Polar): Strongly water-attracting

println("=" ^ 70)
println("HYDROPATHY3: Three-class Hydropathy")
println("=" ^ 70)
reduced_hydro3 = Mycelia.reduce_amino_acid_alphabet(insulin_sequence, :HYDROPATHY3)
info = Mycelia.get_reduced_alphabet_info(:HYDROPATHY3)
println("Description: $(info[:description])")
println("Groups:")
for (letter, aas) in sort(collect(info[:groups]))
    println("  $letter: $aas")
end
println("\nOriginal: $insulin_sequence")
println("Reduced:  $reduced_hydro3")
println()

Part 5: GBMR4 - Isolating Special Amino Acids

The GBMR4 scheme recognizes that Glycine and Proline have unique structural roles:

G (Glycine): Smallest, most flexible
P (Proline): Rigid, breaks secondary structure
H (Hydrophobic): Standard hydrophobic amino acids
B (Basic/polar): Polar and charged amino acids

println("=" ^ 70)
println("GBMR4: Four-class (isolating Gly and Pro)")
println("=" ^ 70)
reduced_gbmr4 = Mycelia.reduce_amino_acid_alphabet(insulin_sequence, :GBMR4)
info = Mycelia.get_reduced_alphabet_info(:GBMR4)
println("Description: $(info[:description])")
println("Groups:")
for (letter, aas) in sort(collect(info[:groups]))
    println("  $letter: $aas")
end
println("\nOriginal: $insulin_sequence")
println("Reduced:  $reduced_gbmr4")
println()

Part 6: Chemical Properties (CHEMICAL6)

The CHEMICAL6 scheme groups by detailed chemical properties:

A (Aliphatic): Non-aromatic hydrophobic chains
R (aRomatic): Contains aromatic rings
+ (Positive): Positively charged (basic)
- (Negative): Negatively charged (acidic)
T (Tiny): Small amino acids
D (Diverse): Other properties

This scheme is particularly useful for studying electrostatic interactions.

println("=" ^ 70)
println("CHEMICAL6: Six-class Chemical Properties")
println("=" ^ 70)
reduced_chem6 = Mycelia.reduce_amino_acid_alphabet(insulin_sequence, :CHEMICAL6)
info = Mycelia.get_reduced_alphabet_info(:CHEMICAL6)
println("Description: $(info[:description])")
println("Groups:")
for (letter, aas) in sort(collect(info[:groups]))
    println("  $letter: $aas")
end
println("\nOriginal: $insulin_sequence")
println("Reduced:  $reduced_chem6")
println()

Part 7: Structure-Dependent Model (SDM12)

The SDM12 scheme (Murphy et al. 2000) maintains the most detail with 12 classes. It preserves important functional groupings while still reducing complexity. This is useful when you need more resolution than HP2 but less than the full 20.

println("=" ^ 70)
println("SDM12: Structure-dependent Model (12 classes)")
println("=" ^ 70)
reduced_sdm12 = Mycelia.reduce_amino_acid_alphabet(insulin_sequence, :SDM12)
info = Mycelia.get_reduced_alphabet_info(:SDM12)
println("Description: $(info[:description])")
println("Groups:")
for (letter, aas) in sort(collect(info[:groups]))
    println("  $letter: $aas")
end
println("\nOriginal: $insulin_sequence")
println("Reduced:  $reduced_sdm12")
println()

Part 8: Comparing All Schemes

Let's see all reductions side-by-side to compare how they simplify the sequence:

println("=" ^ 70)
println("Comparison of All Schemes")
println("=" ^ 70)
println("Original:    $insulin_sequence")
for scheme in Mycelia.list_reduced_alphabets()
    reduced = Mycelia.reduce_amino_acid_alphabet(insulin_sequence, scheme)
    println(rpad("$scheme:", 13), reduced)
end
println()

Part 9: Application - K-mer Pattern Analysis

One practical application of reduced alphabets is simplifying k-mer patterns. By reducing the alphabet, we reduce the number of possible k-mers, making patterns easier to detect and reducing data sparsity.

println("=" ^ 70)
println("Application: Reduced Alphabet K-mer Patterns")
println("=" ^ 70)

# Extract 3-mers from the original sequence
println("Original 3-mers from insulin sequence:")
original_3mers = [String(insulin_sequence[i:i+2]) for i in 1:(length(insulin_sequence)-2)]
println(join(original_3mers, ", "))
println()

# Extract 3-mers from HP2 reduced sequence
println("HP2 reduced 3-mers:")
reduced_seq = Mycelia.reduce_amino_acid_alphabet(insulin_sequence, :HP2)
reduced_3mers = [reduced_seq[i:i+2] for i in 1:(length(reduced_seq)-2)]
println(join(reduced_3mers, ", "))
println()

# Compare pattern complexity
println("Unique 3-mer patterns:")
println("  Original: ", length(unique(original_3mers)), " unique patterns")
println("  HP2:      ", length(unique(reduced_3mers)), " unique patterns")
reduction_percent = round((1 - length(unique(reduced_3mers))/length(unique(original_3mers))) * 100, digits=1)
println("  Reduction: ", reduction_percent, "% fewer patterns")
println()

Part 10: Practical Considerations

When choosing a reduction scheme, consider:

1. Analysis Goals

HP2: Maximum simplification, focus on hydrophobicity
HYDROPATHY3: Balance between simplicity and detail
GBMR4: Important for structure-related analysis
CHEMICAL5/6: When charge matters (e.g., binding sites)
SDM12: When you need more detail but still want reduction

2. Information Loss

More aggressive reductions lose more information
But they also reduce noise and computational complexity
Choose based on signal-to-noise ratio in your data

3. Literature Compatibility

Use schemes from published studies for comparison
Murphy et al. (2000) SDM12 is widely cited
HP2 is the most established for folding studies

Summary

In this tutorial, we learned:

How to convert amino acid sequences to reduced alphabets
The differences between 6 well-established reduction schemes
How reduced alphabets simplify k-mer patterns
When to use different reduction strategies

References

Murphy et al. (2000) Protein Eng. 13(3):149-152 - Original SDM12 paper
Peterson et al. (2009) BMC Bioinformatics 10:228 - Automated reduction methods
Zheng et al. (2019) Database (Oxford) baz131 - RAACBook comprehensive database