Margaret Dayhoff and the Birth of Computational Biology

The field of computational biology, now indispensable for veterinary virology, molecular diagnostics, and systems biology, traces its conceptual and technical origins to the work of Margaret Oakley Dayhoff. Her systematic approach to collecting, comparing, and computing protein sequences established the foundational frameworks for sequence databases, evolutionary models, and alignment algorithms. This article examines Dayhoff's contributions through a technical lens, emphasizing their relevance to modern veterinary molecular diagnostics and pathogen genomics.

The Atlas of Protein Sequence and Structure

Dayhoff's principal achievement was the creation of the Atlas of Protein Sequence and Structure, a compendium that began as a physical collection of all known protein sequences [1]. The first edition, published in 1965, contained approximately 65 sequences. This effort required manual curation from published literature, a process that involved translating one-letter amino acid codes from printed tables into machine-readable formats. The Atlas represented the first systematic attempt to organize biological sequence data, predating the establishment of centralized nucleotide databases by more than a decade [2].

The methodological innovation of the Atlas lay in its use of computational tools for sequence comparison. Dayhoff and her team at the National Biomedical Research Foundation developed early computer programs to align sequences and detect homologous regions. These programs operated on mainframe computers using punched cards as input, a workflow that demanded rigorous error checking and algorithmic efficiency. The computational infrastructure she built directly enabled the discovery of evolutionary relationships among proteins that were not apparent from biochemical characterization alone [1].

Point Accepted Mutation (PAM) Matrices

The most enduring technical contribution from Dayhoff's group is the development of the Point Accepted Mutation (PAM) matrices, also known as Dayhoff matrices [3]. These matrices quantify the probability that one amino acid will be substituted for another over a given evolutionary distance. The construction of PAM matrices required a sophisticated statistical analysis of aligned protein sequences from closely related organisms.

The methodology proceeded as follows. Dayhoff and her colleagues first assembled alignments of proteins with known sequence similarity, typically from different species. They then counted the observed substitutions at each aligned position, normalizing these counts to account for the relative frequencies of each amino acid in the dataset. The resulting substitution probabilities were compiled into a 20x20 matrix, where each entry represented the log-odds ratio of observing a particular substitution relative to random chance.

The PAM1 matrix represented one unit of evolution, defined as the amount of change corresponding to one accepted point mutation per 100 residues. Higher PAM numbers (e.g., PAM250, PAM120) were derived by matrix multiplication, allowing the model to represent increasing evolutionary distances. The PAM250 matrix, which corresponds to approximately 250 substitutions per 100 residues, became widely used for detecting distant evolutionary relationships [3].

The mathematical formulation of PAM matrices can be expressed as follows. For a pair of amino acids i and j, the substitution score S(i,j) is given by:

S(i,j) = log2 ( M(i,j) / f(i) )

where M(i,j) is the probability of amino acid i being substituted by amino acid j per unit evolutionary distance, and f(i) is the background frequency of amino acid i in the dataset. This log-odds formulation ensures that positive scores indicate substitutions occurring more frequently than expected by chance, while negative scores indicate rare substitutions.

The Protein Information Resource

The Atlas evolved into the Protein Information Resource (PIR), an international database of protein sequences [4, 5]. The PIR was established as a collaborative effort between the National Biomedical Research Foundation, the Munich Information Center for Protein Sequences, and the Japan International Protein Sequence Database. This consortium maintained the PIR-International Protein Sequence Database, which provided annotated protein sequences with cross-references to other biological databases.

The PIR database employed a rigorous annotation system that included sequence features, functional domains, and bibliographic references. Each entry was assigned a unique identifier and classified according to a hierarchical system of superfamilies, families, and subfamilies. This classification scheme, based on sequence similarity and evolutionary relationships, enabled researchers to infer the function of newly sequenced proteins by homology to characterized entries [4].

For veterinary diagnostics, the PIR database provided essential reference sequences for pathogen proteins. For example, the identification of surface antigens in Feline Leukemia Virus or capsid proteins in Canine Parvovirus variants relied on comparisons to PIR entries. The database also supported the development of diagnostic assays, such as the Enzyme-Linked Immunosorbent Assay (ELISA) for Feline Leukemia Virus, by providing the sequence information necessary for recombinant antigen production.

Dayhoff's Hypothesis on Protein Evolution

Beyond database construction and matrix development, Dayhoff proposed a hypothesis regarding the origin of functional proteins from short peptides [6]. This hypothesis, which turned 50 years old in 2016, suggested that modern proteins evolved through the duplication, fusion, and diversification of short peptide modules. The hypothesis was based on the observation that many protein sequences contain internal repeats and that these repeats often correspond to functional domains.

Experimental validation of this hypothesis has come from directed evolution studies. Short random peptides, typically 10 to 30 residues in length, can exhibit catalytic activity or binding specificity when properly folded. The Dayhoff hypothesis implies that the earliest proteins were simple peptides that gradually acquired complexity through gene duplication and recombination events. This model has implications for understanding the evolution of virulence factors in veterinary pathogens, where domain shuffling often generates new antigenic variants.

Computational Methods for Sequence Alignment

Dayhoff's work directly enabled the development of modern sequence alignment algorithms. The PAM matrices provided the substitution scores used in dynamic programming algorithms for pairwise alignment, such as the Needleman-Wunsch and Smith-Waterman algorithms. These algorithms require a scoring scheme that rewards matches and penalizes mismatches and gaps. The PAM matrices supplied the biologically meaningful substitution scores that made these alignments accurate for evolutionary inference [3].

The use of PAM matrices in alignment can be illustrated with a simple example. Consider two short peptide sequences from a viral surface protein. The alignment algorithm evaluates all possible alignments, scoring each position using the PAM250 matrix. A match of identical residues receives a positive score, while a substitution such as valine to isoleucine (a conservative change) receives a moderate positive score. A substitution such as valine to glutamic acid (a non-conservative change) receives a negative score. The algorithm then selects the alignment with the highest total score, which corresponds to the most evolutionarily plausible arrangement.

For veterinary applications, this approach is used to compare sequences from field isolates of pathogens such as Highly Pathogenic Avian Influenza (H5N1) in Poultry and Wild Birds. Alignment of hemagglutinin sequences from different isolates reveals amino acid substitutions that may alter antigenicity or host range. The PAM matrices provide the evolutionary context for interpreting these substitutions, distinguishing between neutral drift and adaptive change.

Maximum Likelihood Estimation of Replacement Matrices

The methodology pioneered by Dayhoff has been extended through maximum likelihood estimation of amino acid replacement rate matrices [7]. Modern computational tools, such as the ReplacementMatrix web server, allow researchers to estimate substitution matrices from large sequence alignments using statistical inference. These methods improve upon the original PAM approach by accounting for variation in substitution rates across sites and by using more sophisticated models of sequence evolution.

The maximum likelihood approach estimates the parameters of a continuous-time Markov process describing amino acid substitutions. The rate matrix Q, with entries q(i,j) representing the instantaneous rate of substitution from amino acid i to amino acid j, is estimated by maximizing the likelihood of the observed alignment given the phylogenetic tree. This method produces matrices that are tailored to specific datasets, such as viral proteins evolving under strong selective pressure.

In veterinary virology, these tailored matrices are used to study the evolution of pathogens like West Nile Virus in horses or Feline Coronavirus in domestic cats. The substitution matrices derived from these analyses reveal the selective constraints acting on viral proteins and can identify sites under positive selection, which may correspond to immune evasion or host adaptation.

Legacy and Impact on Veterinary Bioinformatics

Dayhoff's contributions established the conceptual and technical infrastructure for computational biology. The Protein Information Resource, though eventually superseded by UniProt, set the standard for protein sequence databases. The PAM matrices remain in use for sequence alignment and evolutionary analysis, particularly for studies involving divergent sequences where more modern matrices may overfit the data [3].

The practical impact on veterinary diagnostics is substantial. Sequence databases built on Dayhoff's model enable the rapid identification of pathogens through molecular methods. For example, PCR primers for detecting Mycoplasma bovis in Feedlot Cattle are designed by aligning conserved regions of the 16S rRNA gene or species-specific protein genes. The alignment algorithms that underpin this primer design rely on substitution matrices derived from Dayhoff's work.

Similarly, phylogenetic analysis of Escherichia coli in Chickens and Poultry Products uses sequence alignment and evolutionary models to trace the origin of pathogenic strains. The PAM matrices provide the substitution probabilities that allow these analyses to estimate divergence times and transmission routes.

The Dayhoff Workflow for Sequence Analysis

The following Mermaid diagram illustrates the workflow that Dayhoff established for computational sequence analysis, which remains the standard approach in veterinary molecular diagnostics.

flowchart TD
    A[Protein Sequence Collection] --> B[Manual Curation and Digitization]
    B --> C[Sequence Alignment]
    C --> D[Substitution Counting]
    D --> E[PAM Matrix Construction]
    E --> F[Database Entry and Annotation]
    F --> G[Sequence Comparison and Homology Detection]
    G --> H[Functional Inference and Evolutionary Analysis]
    H --> I[Diagnostic Assay Design]
    I --> J[Pathogen Identification and Surveillance]

This workflow begins with the collection of sequence data, proceeds through alignment and statistical analysis, and culminates in practical applications such as diagnostic assay design. Each step in the workflow relies on the computational methods that Dayhoff pioneered.

Conclusion

Margaret Dayhoff's work established the foundations of computational biology through the creation of the Atlas of Protein Sequence and Structure, the development of PAM matrices, and the establishment of the Protein Information Resource. Her contributions enabled the systematic analysis of protein sequences, providing the tools for evolutionary inference, homology detection, and database construction that underpin modern veterinary molecular diagnostics. The methods she developed continue to inform the analysis of pathogen genomes, the design of diagnostic assays, and the study of host-pathogen interactions in veterinary medicine.

References

[1] Strasser BJ. Collecting, comparing, and computing sequences: the making of Margaret O. Dayhoff's Atlas of Protein Sequence and Structure, 1954-1965. J Hist Biol. 2010. URL: https://pubmed.ncbi.nlm.nih.gov/20665074/

[2] Hagen JB. The origin and early reception of sequence databases. Methods Mol Biol. 2011. URL: https://pubmed.ncbi.nlm.nih.gov/21063941/

[3] Mount DW. Using PAM Matrices in Sequence Alignments. CSH Protoc. 2008. URL: https://pubmed.ncbi.nlm.nih.gov/21356854/

[4] George DG, Dodson RJ, Garavelli JS et al. The Protein Information Resource (PIR) and the PIR-International Protein Sequence Database. Nucleic Acids Res. 1997. URL: https://pubmed.ncbi.nlm.nih.gov/9016497/

[5] Barker WC, George DG, Mewes HW et al. The PIR-International databases. Nucleic Acids Res. 1993. URL: https://pubmed.ncbi.nlm.nih.gov/8332528/

[6] Romero Romero ML, Rabin A, Tawfik DS. Functional Proteins from Short Peptides: Dayhoff's Hypothesis Turns 50. Angew Chem Int Ed Engl. 2016. URL: https://pubmed.ncbi.nlm.nih.gov/27865046/

[7] Dang CC, Lefort V, Le VS et al. ReplacementMatrix: a web server for maximum-likelihood estimation of amino acid replacement rate matrices. Bioinformatics. 2011. URL: https://pubmed.ncbi.nlm.nih.gov/21791535/

[8] Palmblad M, Hoopmann MR, Dorfer V. A Special Software Issue in Celebration of Margaret Dayhoff's 100th Birthday. J Proteome Res. 2025. URL: https://pubmed.ncbi.nlm.nih.gov/40051301/

[9] Brannigan V. Margaret Dayhoff: A Personal Memoir. J Proteome Res. 2025. URL: https://pubmed.ncbi.nlm.nih.gov/40051300/

[10] Masic I. The Most Influential Scientists in the Development of Medical informatics (13): Margaret Belle Dayhoff. Acta Inform Med. 2016. URL: https://pubmed.ncbi.nlm.nih.gov/27708497/

[11] Hunt LT. Margaret O. Dayhoff 1925-1983. DNA. 1983. URL: https://pubmed.ncbi.nlm.nih.gov/6347589/

Disclaimer: This article is for educational and informational purposes only. It is not intended to substitute for professional veterinary advice, diagnosis, treatment, or regulatory guidance. Always consult a licensed veterinarian or qualified specialist regarding animal health, disease diagnosis, and therapeutic decisions.