Review: The Rapid Generation of Mutation Data Matrices from Protein Sequences

Citation

• Jones, D. T., Taylor, W. R., & Thornton, J. M. (1992). The rapid generation of mutation data matrices from protein sequences. Bioinformatics (Computer Applications in the Biosciences), 8(3), 275–282.
• DOI

Abstract

A fast method for deriving amino acid substitution matrices from large protein sequence databases is described. The resulting JTT matrix captures the relative rates at which each amino acid is substituted for each other and provides a general empirical model for protein evolution suitable for phylogenetic analysis.

---

The JTT (Jones, Taylor, Thornton) matrix is among the oldest widely used empirical amino acid substitution matrices. It was derived from a database of aligned protein sequences by counting observed substitutions, normalizing for the expected number of substitutions under the null model of equal rates, and symmetrizing the result to satisfy time-reversibility.

The counting approach used in JTT differs from the maximum-likelihood approach used in WAG and LG: rather than finding the substitution rates that maximize the likelihood of the observed data, it tallies observed pairwise replacements in a protein alignment database and converts those counts into rates. This makes the method fast but slightly less statistically efficient than ML estimation.

Despite being superseded by WAG (2001) and LG (2008), JTT remains in active use for three reasons: historical compatibility (many published analyses used JTT, making it the appropriate choice when reproducing or extending earlier work), residual accuracy on some protein families, and as a simple baseline model against which newer matrices are compared.

In Hifuku, JTT is provided as a built-in fixed-rate matrix alongside WAG and LG. The same EigenDecomp class handles all three; only the exchangeability constants and stationary frequencies differ. See WAG and LG for the more accurate alternatives.