# Ancestral Sequence Reconstruction: A Practical Guide to Inferring Ancient Proteins and Testing Evolutionary Hypotheses

Ancestral sequence reconstruction (ASR) is a computational method that infers the likely amino acid sequence of an extinct protein from a multiple sequence alignment of modern descendants and a phylogenetic tree. Researchers use ASR to test hypotheses about how protein function, stability, and specificity evolved. This article provides a practical workflow for conducting ASR experiments, from sequence retrieval through phylogenetic inference, ancestral state estimation, gene synthesis, and functional validation. The guidance is intended for biology students, researchers, and laboratory professionals who need a reproducible protocol with clear decision points and quality controls.

## Scope and Reader Context

ASR sits at the intersection of molecular evolution, structural biology, and biochemistry. The method answers questions such as: What was the catalytic activity of an ancestral enzyme? When did substrate specificity shift? How did protein stability change during adaptation to new environments? The workflow requires competence in sequence database searching, multiple sequence alignment, phylogenetic inference, and ancestral state reconstruction. Each step has software choices, parameter decisions, and validation checks that affect the reliability of the final resurrected protein.

This article covers the complete ASR pipeline with practical guidance for each stage. It assumes familiarity with basic molecular biology and some exposure to command-line tools. The focus is on making defensible choices at each step, documenting those choices, and interpreting results within their limitations. The target reader is a graduate student or postdoctoral researcher planning an ASR project, or a laboratory professional evaluating whether ASR is appropriate for a specific evolutionary question.

## At a Glance

The table below summarizes the main workflow stages, the primary software categories, the key decision points, and the most common sources of error at each step.

| Workflow Stage | Primary Tools | Key Decisions | Common Errors |
| --- | --- | --- | --- |
| Sequence retrieval | NCBI databases, orthology resources | Taxon sampling breadth, sequence quality filtering | Incomplete taxon sampling, inclusion of paralogs |
| Multiple sequence alignment | MAFFT, MUSCLE, PRANK | Alignment algorithm, gap treatment, manual curation | Misaligned variable regions, over-trimming |
| Phylogenetic inference | RAxML, IQ-TREE, BEAST2 | Substitution model, rate heterogeneity, molecular clock | Model misspecification, long-branch attraction |
| Ancestral state reconstruction | PAML, BEAST2, ProtParCon | Marginal versus joint reconstruction, rate model | Ignoring reconstruction uncertainty |
| Gene synthesis and expression | Commercial synthesis, expression vectors | Codon optimization, expression host | Synthesis errors, misfolding, incorrect assay conditions |
| Functional validation | Enzyme assays, binding studies, stability measurements | Assay design, appropriate modern comparators | Overinterpretation of single resurrected variants |

## Understanding the Evolutionary Framework

### The Logic of Ancestral Reconstruction

ASR operates on a simple premise: if a set of modern protein sequences share a common ancestor, and the evolutionary relationships among those sequences can be inferred, then the sequence of the common ancestor can be estimated using a model of sequence evolution. The inference is probabilistic. The software calculates, for each position in the alignment, the probability of each possible amino acid at each internal node of the phylogenetic tree. The researcher then selects either the single most probable amino acid at each position (joint reconstruction) or the set of amino acids with their individual probabilities (marginal reconstruction).

The reliability of the reconstruction depends on three factors. First, the alignment must accurately reflect positional homology. Second, the phylogenetic tree must be a reasonable estimate of the true evolutionary history. Third, the substitution model must capture the actual process of sequence change. Violations of any of these assumptions produce biased ancestral sequences that may not reflect the true ancestral protein.

### What ASR Can and Cannot Reveal

ASR is a hypothesis-generating tool. A reconstructed ancestral sequence represents a statistical estimate, not a certainty. The confidence in any particular residue depends on the number of sequences in the alignment, the evolutionary distance between the ancestor and its descendants, and the conservation of the position. Highly conserved positions are reconstructed with high confidence. Variable positions, particularly those in rapidly evolving regions, may have multiple plausible states.

The resurrected protein, when synthesized and expressed, provides a testable hypothesis about the biochemical properties of an ancient enzyme. If the ancestral protein exhibits a particular substrate preference or stability profile, that observation supports the hypothesis that those properties were present in the ancestor. However, the resurrected protein is a modern construct. It is expressed in a modern host, folded under modern conditions, and assayed with modern substrates. The measured properties may differ from the true ancestral protein due to interactions with other proteins, cellular environment, or post-translational modifications that are not reconstructed.

## Sequence Data Acquisition

### Selecting Sequences for Analysis

The first practical decision is which sequences to include. The choice of taxa determines the phylogenetic scope of the reconstruction and the resolution of ancestral nodes. A reconstruction aimed at the ancestor of a specific enzyme family requires broad sampling across that family, including representatives from all major clades. A reconstruction aimed at a specific node, such as the ancestor of vertebrates, requires sequences from all major vertebrate lineages plus appropriate outgroups.

The [NCBI](https://www.ncbi.nlm.nih.gov/) provides access to comprehensive sequence databases, search systems, and analysis services that support this initial data collection phase. Researchers can search nucleotide and protein databases, retrieve sequences by accession, and access curated reference datasets. The NCBI resources also provide access to the Conserved Domain Database, which can help identify protein families and functional domains within retrieved sequences.

### Orthology and Paralogy Considerations

A critical error in ASR is the inclusion of paralogous sequences. Paralogs are genes that arose through duplication and may have diverged in function. If paralogs are included in a single alignment, the phylogenetic tree will contain two distinct clades, and the ancestral reconstruction will be confounded. The researcher must ensure that all sequences in the alignment are orthologs, meaning they descended from a single ancestral gene.

Orthology assignment can be performed using phylogenetic approaches or through curated databases. The [PAHG database](https://doi.org/10.1186/s12863-025-01361-y) provides curated information on human multigene families, including the phylogenetic history of 221 families with 15,231 protein sequences from diverse metazoans. This resource can help researchers understand duplication events and their timing, which is valuable when deciding whether a gene family has experienced recent duplications that complicate orthology assignment.

### Taxon Sampling Strategy

The number of sequences required depends on the question. A reconstruction of a deep ancestor, such as the ancestral enzyme of a plant flavonoid pathway, benefits from broad sampling across diverse lineages. A [study of 238 seed plant genomes](https://doi.org/10.3389/fpls.2025.1703405) revealed that the F3H, ANS, and FLS enzyme families in flavonoid biosynthesis originated from a common ancestral gene and diversified through multiple duplication events. This level of sampling provides the phylogenetic resolution needed to reconstruct the ancestral enzyme with confidence.

Practical considerations constrain taxon sampling. Sequence availability varies by lineage. Some groups have extensive genomic resources, while others have only a handful of sequences. The researcher should aim for a balanced sampling that includes representatives from all major clades relevant to the question, with outgroup sequences to root the tree. Sparse sampling in critical regions of the tree reduces confidence in the ancestral reconstruction at those nodes.

## Multiple Sequence Alignment

### Alignment Algorithms and Their Tradeoffs

The multiple sequence alignment is the foundation of all downstream analyses. Errors in the alignment propagate through phylogenetic inference and ancestral reconstruction. The choice of alignment algorithm affects the final result, particularly in regions of insertions and deletions.

Popular alignment programs include MAFFT, MUSCLE, and PRANK. MAFFT is fast and accurate for most datasets. MUSCLE offers a good balance of speed and accuracy. PRANK distinguishes between insertion and deletion events, which can improve alignment quality in datasets with many indels but requires more computational time. The choice among these tools depends on the dataset size and the expected pattern of sequence evolution.

The [EMBL-EBI Training portal](https://www.ebi.ac.uk/training) provides learning pathways for bioinformatics analysis, including practical training on sequence alignment and phylogenetic methods. Researchers new to alignment software can use these resources to understand the parameters and their effects on alignment quality.

### Alignment Curation and Trimming

Raw alignments almost always contain regions of questionable homology. Variable loop regions, particularly in proteins with divergent sequences, may align poorly. The researcher must decide whether to trim these regions or retain them with the understanding that they may introduce noise.

Several strategies exist for alignment curation. Manual curation in an alignment editor allows the researcher to inspect each column and remove those that are clearly misaligned. Automated trimming tools remove columns based on gap frequency or conservation thresholds. The choice between manual and automated curation depends on the dataset size and the researcher's familiarity with the protein family.

The decision to trim should be documented. The final alignment, including the trimming criteria, should be archived with the analysis. This documentation allows other researchers to reproduce the analysis and assess the impact of trimming decisions on the results.

### Alignment Quality Assessment

Alignment quality can be assessed through several metrics. The number of conserved columns, the distribution of gaps, and the consistency of the alignment across different algorithms provide indications of reliability. If two different alignment programs produce substantially different alignments, the regions of disagreement should be examined carefully.

The [Galaxy Training Network](https://training.galaxyproject.org/) provides accessible workflow training and analysis tutorials that include guidance on alignment quality assessment. These resources are useful for researchers who want to implement reproducible alignment workflows with documented quality checks.

## Phylogenetic Inference

### Choosing the Inference Method

Phylogenetic inference methods fall into two broad categories: maximum likelihood and Bayesian. Maximum likelihood methods, implemented in programs such as RAxML and IQ-TREE, search for the tree that maximizes the probability of the observed data under a specified substitution model. Bayesian methods, implemented in programs such as BEAST2, sample from the posterior distribution of trees using Markov chain Monte Carlo.

The choice between maximum likelihood and Bayesian methods depends on the question. Maximum likelihood is faster and appropriate for most ASR applications. Bayesian methods provide a distribution of trees, which can be used to incorporate phylogenetic uncertainty into the ancestral reconstruction. If the researcher wants to account for tree uncertainty, Bayesian methods are preferred.

### Substitution Model Selection

The substitution model describes the relative rates of different amino acid replacements. The simplest models assume equal rates for all replacements. More complex models, such as the WAG and LG models, estimate rates from large empirical datasets. The best-fitting model for a particular dataset can be selected using model selection criteria such as the Akaike Information Criterion or the Bayesian Information Criterion.

Model misspecification can bias ancestral reconstruction. If the true substitution process differs substantially from the assumed model, the reconstructed ancestral sequences may be systematically wrong. The researcher should test multiple models and select the one that best fits the data.

### Rate Heterogeneity and Invariant Sites

Proteins evolve at different rates at different positions. Some positions are highly conserved due to functional constraints, while others evolve rapidly. Substitution models that account for rate heterogeneity, typically using a gamma distribution with a proportion of invariant sites, provide a better fit to real protein data than models that assume a uniform rate.

The gamma distribution parameter and the proportion of invariant sites are estimated from the data during phylogenetic inference. These parameters affect branch lengths and, consequently, the ancestral state reconstruction. The researcher should verify that the estimated parameters are reasonable and that the results are not sensitive to small changes in these values.

### Molecular Clock Considerations

The molecular clock assumption states that sequences evolve at a constant rate over time. Strict clocks are rarely appropriate for protein-coding genes, which often show rate variation among lineages. Relaxed clock models allow rates to vary across the tree.

For ASR, the molecular clock is not strictly necessary. The ancestral reconstruction depends on the tree topology and branch lengths, which are estimated without assuming a clock. However, if the researcher wants to date ancestral nodes or correlate ancestral sequences with geological or environmental events, a clock model is required.

The [MASCOT-Skyline approach](https://doi.org/10.1371/journal.pcbi.1013421) demonstrates the importance of modeling temporal and spatial dynamics jointly in phylogeographic analyses. While this work focuses on pathogen spread, the principle applies to ASR: modeling assumptions affect the inferred ancestral states, and joint modeling of relevant processes provides more reliable estimates than modeling each process separately.

## Ancestral State Reconstruction

### Marginal versus Joint Reconstruction

Ancestral state reconstruction methods produce two types of estimates. Marginal reconstruction calculates the probability of each amino acid at each internal node independently, considering all other nodes. Joint reconstruction finds the combination of amino acids across all internal nodes that maximizes the overall probability.

Marginal reconstruction is more commonly used because it provides a measure of uncertainty at each position. The researcher can see which positions are confidently reconstructed and which are ambiguous. Joint reconstruction produces a single sequence but does not provide position-specific confidence estimates.

The [ProtParCon framework](https://pubmed.ncbi.nlm.nih.gov/30813627) provides a common interface for processing molecular data and identifying parallel and convergent amino acid replacements. It integrates multiple sequence alignment, ancestral state reconstruction, maximum-likelihood phylogenetic tree inference, and sequence simulation into a single pipeline. This tool is useful for researchers who want to automate the ASR workflow and identify positions where the same amino acid replacement occurred independently in different lineages.

### Incorporating Phylogenetic Uncertainty

The ancestral reconstruction is conditional on the phylogenetic tree. If the tree is uncertain, the ancestral sequences are uncertain. Bayesian methods can incorporate this uncertainty by averaging the ancestral reconstruction over the posterior distribution of trees.

The practical impact of phylogenetic uncertainty depends on the dataset. For well-resolved trees with strong support at the relevant nodes, the ancestral reconstruction is relatively insensitive to tree uncertainty. For poorly resolved trees, particularly at deep nodes, the ancestral reconstruction may vary substantially across plausible trees.

The researcher should assess the support for the nodes that are ancestral to the sequences of interest. If support is low, the ancestral reconstruction should be interpreted with caution. The [PhyCovA tool](https://pubmed.ncbi.nlm.nih.gov/35295748), while designed for phylogeographic analysis, demonstrates the value of exploring how ancestral state reconstructions change under different models and covariates. Similar exploratory approaches can be applied to ASR to assess the robustness of ancestral sequences to model choices.

### Reconstruction Uncertainty and Ambiguous Positions

Even with a well-supported tree and an appropriate model, some positions in the ancestral sequence will be ambiguous. The marginal reconstruction provides probabilities for each amino acid at each position. A position with a probability of 0.95 for alanine is confidently reconstructed. A position with probabilities of 0.40 for alanine, 0.35 for serine, and 0.25 for threonine is ambiguous.

The researcher must decide how to handle ambiguous positions. One approach is to use the most probable amino acid at each position, accepting that some positions may be wrong. Another approach is to synthesize multiple ancestral variants that differ at ambiguous positions and test them experimentally. This approach, while more expensive, provides a direct assessment of the impact of reconstruction uncertainty on protein function.

## Gene Synthesis and Ancestral Protein Expression

### Sequence Design for Synthesis

Once the ancestral sequence is reconstructed, it must be synthesized as a gene for expression. The reconstructed amino acid sequence is reverse-translated into a DNA sequence using the genetic code. The choice of codons affects expression levels in the host organism.

Codon optimization is the process of adjusting the DNA sequence to match the codon usage preferences of the expression host. Highly expressed genes in a particular organism tend to use a subset of codons. Optimizing the ancestral gene for the expression host can improve protein yields. However, codon optimization can also affect mRNA structure and stability, which may influence expression in unpredictable ways.

The researcher should consider whether to optimize codons or use the codons predicted for the ancestral organism. If the goal is to study the ancestral protein in isolation, codon optimization for the expression host is appropriate. If the goal is to study co-translational folding or other processes that depend on codon usage, the ancestral codons may be preferred.

### Expression System Selection

The choice of expression host depends on the protein and the assays planned. Escherichia coli is the most common host for soluble proteins that do not require extensive post-translational modifications. Yeast systems are useful for proteins that require glycosylation or other eukaryotic modifications. Insect and mammalian cell systems are used for complex eukaryotic proteins.

The expression host affects the measured properties of the resurrected protein. A protein expressed in E. coli may lack post-translational modifications that were present in the ancestral organism. If those modifications affect function, the measured activity may not reflect the true ancestral state. The researcher should consider whether the expression host is appropriate for the protein being studied.

### Purification and Quality Assessment

The resurrected protein must be purified to homogeneity before functional characterization. The purification strategy depends on the protein's properties and the expression system. Affinity tags, such as histidine tags, facilitate initial purification. Additional chromatography steps may be required to remove contaminants.

Protein quality should be assessed before functional assays. SDS-PAGE verifies molecular weight and purity. Mass spectrometry confirms the primary structure and detects post-translational modifications. Circular dichroism spectroscopy assesses secondary structure. If the protein is misfolded or aggregated, the functional assays will produce misleading results.

## Functional Characterization of Ancestral Proteins

### Assay Design and Controls

The functional assays used to characterize the resurrected protein should be designed to test the specific evolutionary hypothesis. If the hypothesis concerns substrate specificity, the assay should measure activity against multiple substrates. If the hypothesis concerns thermal stability, the assay should measure activity or structure at multiple temperatures.

Modern comparators are essential controls. The resurrected protein should be assayed alongside extant proteins from the same family. These comparisons provide context for interpreting the ancestral protein's properties. If the ancestral protein is more stable than all modern proteins, that observation supports the hypothesis that stability decreased over evolutionary time. If the ancestral protein is similar to modern proteins, the hypothesis is not supported.

### Biophysical Characterization

Beyond enzymatic activity, biophysical properties provide insight into protein evolution. Thermal stability, measured by circular dichroism or differential scanning fluorimetry, reveals the melting temperature of the ancestral protein. Kinetic stability, measured by resistance to proteolysis or denaturation, reflects the protein's resistance to unfolding.

These measurements are particularly informative for testing hypotheses about environmental adaptation. If an ancestral protein is inferred to have existed in a hot environment, its thermal stability should be higher than that of modern proteins from cooler environments. The biophysical data provide a direct test of such hypotheses.

### Structural Characterization

Structural information can be obtained experimentally or computationally. X-ray crystallography and cryo-electron microscopy provide high-resolution structures of the resurrected protein. These structures can be compared to modern structures to identify structural changes that accompanied functional evolution.

Computational structure prediction provides an alternative when experimental structures are not available. The predicted structure of the ancestral protein can be compared to predicted or experimental structures of modern proteins. These comparisons can reveal changes in active site geometry, substrate binding pockets, or protein-protein interaction surfaces.

The interpretation of structural data should consider the limitations of the methods. Computational predictions are models, not experimental observations. The confidence in predicted structures varies by region, with well-predicted cores and less reliable loops. The researcher should assess the confidence of the prediction before drawing conclusions about specific structural features.

## Records and Documentation

### Reproducibility Requirements

ASR analyses must be reproducible. The complete workflow, from sequence retrieval through ancestral reconstruction, should be documented with sufficient detail that another researcher can repeat the analysis and obtain the same results. This documentation includes software versions, parameter settings, input files, and output files.

The [nf-core documentation](https://nf-co.re/docs) provides standards for community pipeline usage, configuration, and reproducible workflow context. While nf-core pipelines are designed for high-throughput sequencing analysis, the principles of reproducibility apply to ASR workflows. Version control, containerization, and automated pipeline execution ensure that analyses can be reproduced exactly.

### Data Archiving

The input data, intermediate files, and final results should be archived in a publicly accessible repository. Sequence alignments, phylogenetic trees, and ancestral sequences should be deposited in appropriate databases. The specific repository depends on the data type and the journal requirements.

The [NCBI](https://www.ncbi.nlm.nih.gov/) provides data resources for sequence and phylogenetic data. Researchers can deposit alignments and trees in appropriate databases and provide accession numbers in publications. This archiving ensures that the data are available for verification and reanalysis.

### Analysis Logs

A detailed analysis log records every step of the workflow. The log includes the date of each analysis, the software version, the command used, and the output files generated. This log serves as the primary record of the analysis and is essential for troubleshooting and for responding to reviewer questions.

The [Carpentries lessons](https://carpentries.org/lessons) provide foundational training in computing, data, shell, Git, and programming. These skills are directly applicable to ASR workflows. Version control with Git allows the researcher to track changes to analysis scripts and document the evolution of the analysis.

## Common Failure Patterns

### Inadequate Taxon Sampling

The most common failure in ASR is inadequate taxon sampling. If the alignment contains too few sequences, or if the sequences are concentrated in a few closely related lineages, the ancestral reconstruction will be unreliable. The researcher should assess whether the sampling covers the diversity of the protein family and includes appropriate outgroups.

The solution is to expand the sampling. Additional sequences can be retrieved from public databases. The [NCBI](https://www.ncbi.nlm.nih.gov/) provides search systems that support comprehensive sequence retrieval. The researcher should aim for a sampling that includes multiple representatives from each major clade.

### Alignment Artifacts

Alignment errors are a frequent source of biased ancestral reconstructions. Misaligned regions introduce spurious substitutions that distort branch lengths and ancestral states. The researcher should inspect the alignment carefully, particularly in variable regions and regions with indels.

The solution is to improve the alignment. Different alignment algorithms may produce different results. Comparing alignments from multiple programs and examining the regions of disagreement can identify problematic regions. Manual curation may be necessary for difficult regions.

### Model Misspecification

The substitution model may not capture the true process of sequence evolution. Proteins with strong functional constraints, such as active site residues, evolve differently from unconstrained regions. Models that do not account for these differences may produce biased reconstructions.

The solution is to test multiple models and assess the sensitivity of the results to model choice. If the ancestral reconstruction changes substantially under different models, the results should be interpreted with caution. The researcher should report the model selection procedure and the sensitivity analysis.

### Overinterpretation of Results

The final failure pattern is overinterpretation. A resurrected protein that exhibits a particular property in vitro does not prove that the ancestral protein had that property in vivo. The reconstruction is a statistical estimate, the expression system is modern, and the assay conditions are artificial.

The solution is to frame conclusions appropriately. The results support or refute specific hypotheses within the limitations of the method. The researcher should acknowledge the uncertainties and discuss alternative interpretations.

## Limitations and Interpretation Boundaries

### Statistical Uncertainty

The ancestral sequence is a statistical estimate with position-specific uncertainty. The researcher should report the posterior probabilities or bootstrap support for the reconstructed residues. Positions with low support should be identified, and the impact of this uncertainty on the conclusions should be discussed.

### Evolutionary Model Assumptions

All ASR methods assume that the substitution model accurately describes the evolutionary process. This assumption is rarely fully satisfied. Proteins may experience shifts in selective pressure, changes in substitution rates, or episodes of positive selection that are not captured by the model. The researcher should consider whether these processes are relevant to the protein family being studied.

### Modern Context of Resurrected Proteins

The resurrected protein is expressed and assayed in a modern context. The ancestral protein existed in an ancient cellular environment with different temperatures, pH, interacting partners, and substrate concentrations. The measured properties may not reflect the properties of the protein in its ancestral context.

The researcher should consider whether the assay conditions are appropriate for testing the evolutionary hypothesis. If the hypothesis concerns adaptation to a specific environment, the assays should be designed to reflect that environment as closely as possible.

### The PAHG Database as a Cautionary Example

The [PAHG database](https://doi.org/10.1186/s12863-025-01361-y) illustrates the complexity of gene family evolution. The database details the phylogenetic history of 221 human multigene families with 15,231 protein sequences from diverse metazoans, providing insights into gene duplication timings and their relationships with genome organization. This complexity means that simple assumptions about orthology and gene family evolution may be incorrect. The researcher should verify that the gene family being studied does not have a complex duplication history that complicates ancestral reconstruction.

## Professional Escalation Criteria

### When to Seek Expert Assistance

Certain situations warrant consultation with a bioinformatics specialist or a molecular evolution expert. These situations include:

- The protein family has a complex duplication history with multiple paralogs that are difficult to distinguish.
- The phylogenetic tree has poorly resolved nodes at the positions where ancestral reconstruction is needed.
- The alignment contains large regions of uncertain homology that resist automated or manual curation.
- The ancestral reconstruction produces sequences with many ambiguous positions that substantially affect the experimental results.
- The functional assays produce results that are inconsistent with the evolutionary hypothesis and the inconsistency cannot be explained by obvious experimental artifacts.

### When to Reconsider the Project Design

Some situations indicate that the ASR approach may not be appropriate for the question. These situations include:

- The protein family has experienced extensive gene loss, making it impossible to sample the full diversity of the ancestral lineage.
- The sequences are so divergent that alignment is unreliable across most of the protein.
- The protein has experienced strong positive selection that is not captured by standard substitution models.
- The functional properties of interest depend on interactions with other proteins that cannot be reconstructed.

In these situations, the researcher should consider alternative approaches. Comparative analysis of modern proteins, directed evolution, or biophysical characterization of extant proteins may provide more reliable insights than ASR.

## Building a Decision Framework for ASR Project Planning

Before collecting sequences or running alignment software, researchers benefit from a structured decision framework that maps the evolutionary question to specific methodological choices. This framework reduces the risk of investing months in a reconstruction that cannot answer the intended question. The framework operates at three levels: defining the target node, assessing data feasibility, and selecting the validation strategy.

### Defining the Target Node and Its Requirements

The first decision is identifying the exact node on the phylogeny that represents the ancestor of interest. This choice determines the taxon sampling strategy, the required phylogenetic resolution, and the confidence thresholds for the reconstruction. A researcher studying the ancestral enzyme of a plant flavonoid pathway must decide whether the target is the ancestor of all seed plants, the ancestor of a specific family, or the ancestor of a particular clade within that family. Each target node requires different sampling.

The target node should be defined in writing before data collection begins. The definition should specify the taxonomic scope, the expected divergence time if known, and the specific functional hypothesis to be tested. This written definition serves as the reference point for all subsequent decisions. If the target node is deep, such as the ancestor of a large enzyme family, the researcher should expect lower reconstruction confidence at variable positions and plan for multiple ancestral variants to test the impact of ambiguity.

The [PAHG database](https://doi.org/10.1186/s12863-025-01361-y) provides a practical example of how gene family complexity affects target node definition. The database documents the phylogenetic history of 221 human multigene families, revealing that gene duplications occurred at different times and through different mechanisms. A researcher studying a gene family with a complex duplication history must decide whether the target node predates or postdates specific duplication events. This decision changes the set of sequences that should be included in the alignment.

### Assessing Data Feasibility Before Committing to the Project

The second decision point is a feasibility assessment based on available sequence data. This assessment should occur before any alignment or phylogenetic analysis. The researcher needs to determine whether sufficient sequences exist across the taxonomic range required for the target node. A quick search of the [NCBI](https://www.ncbi.nlm.nih.gov/) databases can reveal the number of available sequences for the protein family of interest across relevant lineages.

The feasibility assessment should answer three questions. First, are there enough sequences from the major clades that descend from the target node? Second, are there appropriate outgroup sequences to root the tree? Third, is the sequence divergence within the family manageable for alignment and reconstruction? If the answer to any of these questions is no, the researcher should either adjust the target node or reconsider whether ASR is appropriate.

A practical approach is to retrieve all available sequences for the protein family, remove obvious redundancies and partial sequences, and then map the remaining sequences onto a rough taxonomic tree. This mapping reveals gaps in sampling. If critical clades are missing, the researcher can search for additional sequences in less obvious sources, such as transcriptome assemblies or genome annotations that may not appear in standard protein database searches.

### Selecting the Validation Strategy

The third decision in the framework is the validation strategy. This decision determines the experimental work required after the ancestral sequence is reconstructed. The validation strategy should be chosen before synthesis, because it affects the number of ancestral variants to synthesize and the assays to perform.

Three validation strategies are commonly used. The first is single-variant validation, where one ancestral sequence is synthesized and characterized. This strategy is appropriate when the reconstruction confidence is high across most positions and the functional hypothesis makes specific predictions about the ancestral protein's properties. The second is multi-variant validation, where multiple ancestral sequences that differ at ambiguous positions are synthesized and compared. This strategy is appropriate when reconstruction uncertainty is concentrated in positions that may affect function. The third is comparative validation, where the ancestral protein is characterized alongside a panel of modern proteins from the same family. This strategy is always recommended but is essential when the hypothesis concerns changes in function or stability over evolutionary time.

The [ProtParCon framework](https://pubmed.ncbi.nlm.nih.gov/30813627) illustrates how computational tools can support the validation strategy. ProtParCon integrates multiple sequence alignment, ancestral state reconstruction, maximum-likelihood phylogenetic tree inference, and sequence simulation into a single pipeline. This integration allows researchers to identify parallel and convergent amino acid replacements, which are positions where the same replacement occurred independently in different lineages. These positions are particularly informative for validation because they suggest positions where function may depend on specific amino acid identities.

### Recording Decisions in a Project Log

The decision framework should be documented in a project log that records the rationale for each choice. The log should include the target node definition, the feasibility assessment results, the taxon sampling strategy, and the validation strategy. This log serves multiple purposes. It provides a reference for troubleshooting when results are unexpected. It documents the decision process for reviewers and readers. It also provides a template for future ASR projects in the same laboratory.

The [Carpentries lessons](https://carpentries.org/lessons) provide foundational training in computing, data, shell, Git, and programming that supports this documentation process. Version control with Git allows the researcher to track changes to analysis scripts and documentation over time. The project log should be maintained in a version-controlled repository alongside the analysis scripts and data files.

### Common Decision Errors and Their Consequences

Several decision errors recur across ASR projects. The first is choosing a target node that is too deep for the available data. If the target node is ancient and the descendant sequences are highly divergent, the reconstruction confidence will be low at many positions. The consequence is an ancestral sequence with many ambiguous positions that may not accurately represent the true ancestor.

The second error is selecting sequences without a clear orthology assessment. Including paralogs in the alignment produces a tree with multiple clades that do not represent the evolutionary history of a single gene. The consequence is an ancestral reconstruction that represents a chimeric sequence from multiple gene lineages. The [PAHG database](https://doi.org/10.1186/s12863-025-01361-y) provides curated information on gene duplication events that can help researchers avoid this error for human gene families.

The third error is choosing a validation strategy that cannot distinguish between competing hypotheses. If the hypothesis predicts a change in substrate specificity, the validation must include assays with multiple substrates. If the validation only measures activity with a single substrate, the results cannot test the hypothesis. The consequence is an experiment that produces data but does not answer the question.

The fourth error is failing to assess the sensitivity of the reconstruction to model choices. The ancestral sequence may change substantially when different substitution models or alignment algorithms are used. The consequence is a false sense of confidence in a reconstruction that is not robust to reasonable methodological variation. The [PhyCovA tool](https://pubmed.ncbi.nlm.nih.gov/35295748), while designed for phylogeographic analysis, demonstrates the value of exploring how ancestral state reconstructions change under different models and covariates. Similar exploratory approaches can be applied to ASR to assess the robustness of ancestral sequences to model choices.

### Implementing the Framework in Practice

The decision framework translates into a specific sequence of actions. First, write the target node definition and the functional hypothesis. Second, search the [NCBI](https://www.ncbi.nlm.nih.gov/) databases for available sequences and assess feasibility. Third, select the taxon sampling strategy based on the feasibility assessment. Fourth, choose the validation strategy and determine how many ancestral variants will be needed. Fifth, document all decisions in the project log.

This framework does not guarantee a successful reconstruction, but it reduces the risk of common failures. The framework forces the researcher to make explicit decisions about the target, the data, and the validation before investing in the analysis. These decisions can be revisited as the project progresses, but having them documented from the start provides a clear reference point for evaluating whether the project is on track.

The [Galaxy Training Network](https://training.galaxyproject.org/) provides accessible workflow training and analysis tutorials that can help researchers implement this framework in practice. The tutorials cover sequence retrieval, alignment, phylogenetic inference, and related topics in a reproducible format. Researchers can adapt these workflows to their specific projects and integrate them with the decision framework described here.

The [EMBL-EBI Training portal](https://www.ebi.ac.uk/training) offers learning pathways for bioinformatics analysis that complement the practical workflow implementation. These resources help researchers understand the underlying principles of the methods they are using, which supports better decision making throughout the project.

## Frequently Asked Questions

### What is the minimum number of sequences needed for ancestral sequence reconstruction?

There is no fixed minimum number, but the reliability of the reconstruction increases with the number of sequences and the breadth of taxonomic sampling. A reconstruction based on fewer than ten sequences is generally unreliable because the phylogenetic tree will be poorly resolved and the substitution model parameters will be poorly estimated. A practical target is at least 20 to 50 sequences spanning the diversity of the protein family, with multiple representatives from each major clade. The specific number depends on the evolutionary distance among the sequences and the conservation of the protein.

### How do I choose between maximum likelihood and Bayesian methods for ancestral reconstruction?

Maximum likelihood methods are faster and appropriate for most applications. They produce a single tree and a single set of ancestral reconstructions. Bayesian methods sample from the posterior distribution of trees and can incorporate phylogenetic uncertainty into the ancestral reconstruction. If the tree is well resolved and the researcher wants a straightforward analysis, maximum likelihood is sufficient. If the tree has uncertain regions or the researcher wants to account for tree uncertainty, Bayesian methods are preferred.

### What is the difference between marginal and joint ancestral reconstruction?

Marginal reconstruction calculates the probability of each amino acid at each internal node independently. It provides a measure of uncertainty at each position. Joint reconstruction finds the combination of amino acids across all internal nodes that maximizes the overall probability. Marginal reconstruction is more commonly used because it provides position-specific confidence estimates that help the researcher identify ambiguous positions.

### Should I codon optimize the ancestral gene for expression in my host system?

Codon optimization can improve expression levels by matching the codon usage preferences of the expression host. However, codon optimization can also affect mRNA structure and stability. If the goal is to study the ancestral protein in isolation, codon optimization for the expression host is appropriate. If the goal is to study processes that depend on codon usage, such as co-translational folding, the ancestral codons may be preferred. The decision should be documented and justified.

### How do I handle ambiguous positions in the ancestral reconstruction?

Ambiguous positions are those where the posterior probability is distributed across multiple amino acids. One approach is to use the most probable amino acid at each position, accepting that some positions may be incorrect. Another approach is to synthesize multiple ancestral variants that differ at ambiguous positions and test them experimentally. The second approach is more expensive but provides a direct assessment of the impact of reconstruction uncertainty on protein function.

### What controls should I include in functional assays of resurrected proteins?

Modern comparators are essential controls. The resurrected protein should be assayed alongside extant proteins from the same family under identical conditions. These comparisons provide context for interpreting the ancestral protein's properties. Additional controls include a known positive control for the assay, a negative control without enzyme, and replicates to assess experimental variability.

### How do I assess the reliability of my ancestral reconstruction?

Reliability is assessed through multiple lines of evidence. The posterior probabilities or bootstrap support for reconstructed residues indicate statistical confidence. Sensitivity analysis, in which the reconstruction is repeated with different alignments, models, or taxon sets, reveals the robustness of the results. Experimental validation, in which the resurrected protein is characterized and compared to predictions, provides the ultimate test of the reconstruction.

### What are the main limitations of ancestral sequence reconstruction?

The main limitations are statistical uncertainty, model assumptions, and the modern context of resurrected proteins. The ancestral sequence is a statistical estimate with position-specific uncertainty. The substitution model may not capture the true evolutionary process. The resurrected protein is expressed and assayed in a modern context that may differ from the ancestral environment. These limitations should be acknowledged and discussed in any report of ASR results.

## Related Bioinformatics Guides

- [Ancestral Sequence Reconstruction for Viral Evolution](/knowledge/bioinformatics/ancestral-sequence-reconstruction-for-viral-evolution)
- [Functional Metagenomics: From Gene Prediction to Pathway Reconstruction](/knowledge/bioinformatics/functional-metagenomics-from-gene-prediction-to-pathway-reconstruction)
- [Gene Set Enrichment Analysis in R: A Practical Tutorial for Interpreting Omics Data](/knowledge/bioinformatics/gene-set-enrichment-analysis-in-r-a-practical-tutorial-for-interpreting-omics-data)
- [Metabolomics Data Analysis in R: A Practical Workflow](/knowledge/bioinformatics/metabolomics-data-analysis-in-r-a-practical-workflow)
- [Metagenomics Tools: A Practical Guide to Software and Pipelines](/knowledge/bioinformatics/metagenomics-tools-a-practical-guide-to-software-and-pipelines)

## References and Further Reading

- [NCBI Data Resources](https://www.ncbi.nlm.nih.gov/). National Center for Biotechnology Information.
- [EMBL-EBI Training](https://www.ebi.ac.uk/training). European Bioinformatics Institute.
- [Bioconductor](https://bioconductor.org/). Bioconductor Project.
- [Galaxy Training Network](https://training.galaxyproject.org/). Galaxy Project.
- [nf-core Documentation](https://nf-co.re/docs). nf-core.
- [The Carpentries Lessons](https://carpentries.org/lessons). The Carpentries.
- [Phycova - a tool for exploring covariates of pathogen spread.](https://pubmed.ncbi.nlm.nih.gov/35295748). Virus evolution, 2022.
- [ProtParCon: A Framework for Processing Molecular Data and Identifying Parallel and Convergent Amino Acid Replacements.](https://pubmed.ncbi.nlm.nih.gov/30813627). Genes, 2019.
- [MASCOT-Skyline integrates population and migration dynamics to enhance phylogeographic reconstructions.](https://doi.org/10.1371/journal.pcbi.1013421). 2025.
- [PAHG: the database of human multi-gene families.](https://doi.org/10.1186/s12863-025-01361-y). 2025.
- [Genomic analyses of 238 seed plants reveal the evolutionary mechanisms driving specialization of F3H, ANS, and FLS in flavonoid biosynthesis.](https://doi.org/10.3389/fpls.2025.1703405). 2025.

> This article is educational and does not replace validated analysis plans, institutional policy, clinical interpretation, or specialist review.