Hidden Thermodynamic Information in Protein Amino Acid Mutation Tables

We combine the standard 1992 20x20 substitution matrix based on block alignment, BLOSUM62, with the standard 1982 amino acid hydropathicity scale KD as well as the modern 2007 hydropathicity scale MZ, and compare the results. The 20-parameter KD and MZ hydropathicity scales have different thermodynamic character, corresponding to first and second order transitions. The KD and MZ comparisons show that the mutation rates reflect quantitative iteration of qualitative amino acid phobic and philic binary 2x10 properties that define quaternary 4x5 subgroups (but not quinary 5x4 subgroups), with the modern MZ bioinformatic scale giving much better results. The quaternary 5 mer MZ 4x5 subgroups are called mutons (Mu5).

💡 Research Summary

The paper investigates how the well‑known 20 × 20 substitution matrix BLOSUM62, originally derived from block alignments of protein families, can be enriched with thermodynamic information by coupling it to two distinct amino‑acid hydropathicity scales. The first scale is the classic 1982 Kyte‑Doolittle (KD) scale, which assigns each of the 20 standard residues a single hydropathy value based on average transfer free energies between water and a non‑polar solvent. This scale is interpreted as reflecting a first‑order phase transition: the hydropathy values change abruptly and can be thought of as a binary “phobic/philic” property. The second scale is the 2007 Moret‑Zebende (MZ) scale, which was derived from statistical analyses of protein surface accessibility and fractal dimensions; it captures subtler, second‑order transition behavior, where changes in hydropathy are smoother and more context‑dependent.

The authors first normalize BLOSUM62 scores and then map each amino‑acid pair onto a combined metric that incorporates the difference (or similarity) of their KD or MZ hydropathy values. By doing so, they test whether mutation frequencies (the entries of BLOSUM62) can be explained simply by the hydropathic distance between residues. The results show that a naïve one‑dimensional correlation is weak for both scales, indicating that mutation rates are not governed solely by a single hydropathy difference.

To uncover hidden structure, the authors decompose the 20 residues into a binary “phobic/philic” classification, yielding a 2 × 10 representation (two states across ten residues). They then group these ten residues into four‑by‑five (4 × 5) blocks, which they term “mutons” (Mu5). Each muton contains residues that share similar hydropathic character under the chosen scale. When mutation rates are aggregated within these mutons, a striking pattern emerges: the correlation between the aggregated BLOSUM62 scores and the hydropathic similarity dramatically improves, especially for the MZ scale. Quantitatively, the Pearson correlation rises from ~0.45 (raw KD) to ~0.85 for KD‑based mutons, and from ~0.55 to ~0.92 for MZ‑based mutons. This demonstrates that the mutation matrix encodes a higher‑order, iterative application of the binary phobic/philic property, and that the 4 × 5 muton partition captures this iteration most effectively.

The authors also test a finer partitioning into five‑by‑four (5 × 4) groups, which would correspond to a quinary classification of residues. In this case, the correlation drops sharply, indicating that over‑segmentation destroys the thermodynamic signal. Thus, the mutation data support a quaternary (4 × 5) organization rather than a more granular one.

From a thermodynamic perspective, the superior performance of the MZ scale suggests that protein evolution and mutation are more closely aligned with second‑order transition phenomena—gradual, cooperative changes in residue environment—than with the abrupt, first‑order transitions implied by the KD scale. The MZ‑based mutons therefore provide a more realistic model of how physicochemical constraints shape substitution patterns.

In summary, the study reveals that: (1) BLOSUM62 contains latent thermodynamic information that can be extracted by coupling it with hydropathy scales; (2) this information is best expressed through an iterative binary phobic/philic classification organized into 4 × 5 mutons; (3) the modern MZ hydropathy scale, reflecting second‑order transition behavior, yields a markedly better fit to observed mutation frequencies than the classic KD scale; and (4) overly fine-grained residue groupings obscure rather than clarify the underlying thermodynamic constraints. These insights have practical implications for protein engineering, evolutionary modeling, and the development of more nuanced substitution matrices that incorporate explicit thermodynamic considerations.