# A Decision Guide to Protein Domain Databases: Pfam, SCOP, CATH, and InterPro - Which One Should You Use?

Protein domain databases answer different questions about protein structure, evolution, and function. Pfam organizes domains by sequence similarity, SCOP and CATH classify domains by three-dimensional structure, and InterPro integrates multiple signature databases into a unified search system. For a researcher deciding which database to use, the practical answer depends on whether the starting material is a sequence, a structure, or a set of functional annotations. Sequence-based questions about conserved regions and family membership point to Pfam or InterPro. Structure-based questions about fold classification and evolutionary relationships point to SCOP or CATH. Questions that combine both, such as interpreting molecular docking results or validating predicted structures, often require cross-referencing multiple databases. This guide provides a decision framework for matching research questions to the appropriate database, with attention to data inputs, workflow choices, quality checks, and interpretation limits.

## The Core Problem: Matching Database Design to Research Questions

Protein domain databases are not interchangeable tools. Each database was built with a different classification philosophy, uses different input data, and produces different types of output. Pfam builds multiple sequence alignments and profile hidden Markov models from protein families. SCOP and CATH start from experimentally determined three-dimensional structures and classify domains by structural and evolutionary criteria. InterPro aggregates signatures from Pfam and other member databases, providing a single entry point for functional annotation. The choice of database changes what can be concluded from an analysis, and using the wrong database for a question can produce misleading results.

The practical consequence is that researchers need to understand the underlying logic of each database before selecting one. A sequence that returns no Pfam hit may still have a known structural fold in CATH. A structural classification in SCOP may not correspond to a sequence family in Pfam. The databases are complementary, and the decision framework presented here treats them as such.

## At a Glance: Database Comparison for Common Research Tasks

| Research Task | Recommended Database | Input Required | Output Type | Best Used When |
| --- | --- | --- | --- | --- |
| Identify conserved sequence domains in a novel protein | Pfam or InterPro | Protein sequence | Domain family assignments, functional annotations | Starting with sequence data and needing family membership or functional clues |
| Classify a protein structure by fold and evolutionary family | SCOP or CATH | Protein structure (PDB format) | Structural class, fold, superfamily, family assignments | Working with experimentally determined or confidently predicted structures |
| Combine multiple annotation sources for genome-scale analysis | InterPro | Protein sequence or genome annotation | Integrated functional signatures from multiple member databases | Running large-scale annotation pipelines or needing standardized outputs |
| Compare domain architectures across species | Pfam domain assignments | Protein sequences from multiple organisms | Domain architecture comparisons, co-occurrence patterns | Studying multidomain proteins and domain shuffling |
| Interpret molecular docking results or validate predicted structures | SCOP or CATH with cross-referencing | Protein structure | Structural context for binding sites and domain interfaces | Evaluating whether predicted or docked complexes place domains in plausible arrangements |

## Understanding the Four Databases

### Pfam: Sequence-Based Domain Families

Pfam classifies protein domains by sequence similarity. The database builds multiple sequence alignments for each family and uses these alignments to construct profile hidden Markov models. These models capture the conserved sequence patterns of a domain family and can search new sequences for matches. Pfam families are organized hierarchically, with related families grouped into clans. The database provides functional annotations for many families, including descriptions of domain function and links to other resources.

The strength of Pfam is its sensitivity to distant sequence relationships. Profile hidden Markov models can detect homology that pairwise sequence alignment methods miss, particularly for sequences that have diverged substantially. This makes Pfam useful for annotating novel sequences and for identifying conserved regions that may correspond to functional domains. The [NCBI Data Resources](https://www.ncbi.nlm.nih.gov/) provide access to sequence databases and search tools that complement Pfam analysis, allowing researchers to retrieve sequences and perform initial similarity searches before submitting them to Pfam.

Pfam has limitations. Sequence-based classification does not directly capture structural information, and domains that share a fold but have diverged beyond sequence recognition may not be grouped together. Pfam also struggles with some types of repetitive or low-complexity regions. For structural questions, Pfam assignments need to be interpreted alongside structural classifications.

### SCOP: Structural Classification by Evolutionary Relationships

SCOP classifies protein domains from experimentally determined structures according to structural and evolutionary relationships. The classification hierarchy proceeds from class to fold to superfamily to family. Class describes the secondary structure composition, fold describes the overall arrangement of secondary structure elements, superfamily groups domains with probable common evolutionary origin, and family groups domains with clear sequence similarity.

The SCOP classification emphasizes evolutionary relationships. Domains placed in the same superfamily are inferred to share a common ancestor, even if their sequences have diverged considerably. This makes SCOP valuable for studying protein evolution and for identifying distant structural relationships that sequence-based methods would miss. The structural similarity detection methods used in databases like the Molecular Modeling Database and VAST+ support this kind of analysis by computing large-scale structural alignments and identifying similar domains across the protein structure archive, as described in the [MMDB and VAST+ study](https://pubmed.ncbi.nlm.nih.gov/24319143).

SCOP requires a three-dimensional structure as input. This limits its direct applicability to proteins without experimental structures, although predicted structures can be classified if they are reliable. The classification is also curated, which means it reflects expert judgment about evolutionary relationships instead of purely automated clustering.

### CATH: Structural Classification by Automated and Manual Curation

CATH classifies protein domains from experimentally determined structures using a combination of automated and manual methods. The name reflects the four main levels of the hierarchy: Class, Architecture, Topology, and Homologous superfamily. Class describes secondary structure composition, Architecture describes the arrangement of secondary structure elements without considering sequence order, Topology describes the overall fold with sequence order considered, and Homologous superfamily groups domains with evidence of common ancestry.

CATH places more emphasis on automated classification than SCOP, although manual curation remains important. The database is updated regularly as new structures are deposited. CATH also provides additional resources, including analyses of domain interfaces and functional annotations. The [domain interface study](https://pubmed.ncbi.nlm.nih.gov/10708645) used CATH classifications to analyze the physical and chemical properties of domain-domain interactions, demonstrating the utility of CATH for studying how domains interact within proteins.

CATH and SCOP often agree on broad classifications but differ in details. The two databases use different criteria for defining domain boundaries and for grouping domains into superfamilies. Researchers should not assume that a SCOP classification and a CATH classification will be identical for the same structure.

### InterPro: Integrated Functional Annotation

InterPro integrates protein signature databases into a single resource. Member databases include Pfam, as well as databases based on different types of signatures such as PROSITE patterns, PRINTS fingerprints, and PANTHER families. InterPro searches all member databases simultaneously and provides a unified result that shows which signatures match a given sequence.

The practical advantage of InterPro is convenience and consistency. A single search can identify matches from multiple databases, and the results are presented in a standardized format. This is particularly useful for genome-scale annotation, where running each member database separately would be time-consuming and would produce outputs that are difficult to compare. The [EMBL-EBI Training](https://www.ebi.ac.uk/training) resources provide instruction on using InterPro and other EBI databases for sequence analysis, which can help researchers design effective annotation workflows.

InterPro does not replace the individual databases. The integrated results can obscure differences in how member databases define domains, and researchers who need detailed information about a specific family may need to consult the source database. InterPro is best understood as a front end that provides broad coverage and standardized output, with the individual databases providing depth.

## Practical Workflow: Selecting a Database for Your Research Question

### Step 1: Define the Starting Material

The first decision point is whether the analysis starts with a protein sequence or a protein structure. Sequences are the most common starting point for researchers working with newly discovered genes, predicted proteins from genome assemblies, or proteins identified in proteomics experiments. Structures are the starting point for researchers working with crystallography, cryo-electron microscopy, or nuclear magnetic resonance data, as well as for researchers evaluating predicted structures from tools like AlphaFold.

For sequence inputs, Pfam and InterPro are the primary options. Pfam provides detailed family information and profile hidden Markov model searches. InterPro provides integrated annotation from multiple databases. For structure inputs, SCOP and CATH are the primary options. Both require a structure file in PDB format or a PDB identifier.

### Step 2: Identify the Research Question

The research question determines which database features are most relevant. Questions about conserved sequence regions and family membership point to Pfam. Questions about integrated functional annotation and genome-scale analysis point to InterPro. Questions about structural fold and evolutionary relationships point to SCOP or CATH. Questions about domain architecture and domain combinations point to Pfam domain assignments analyzed with specialized tools.

The [domain architecture comparison study](https://pubmed.ncbi.nlm.nih.gov/19958515) illustrates how Pfam domain assignments can be used to compare proteins at the domain architecture level. The study developed a weighted measure of domain architecture similarity that accounts for promiscuous domains, which appear in many different protein contexts. This approach is useful for identifying homology among multidomain proteins where sequence similarity alone is insufficient.

### Step 3: Consider the Need for Cross-Referencing

Many research questions require information from multiple databases. A researcher studying a novel protein might start with an InterPro search to get broad functional annotation, then consult Pfam for detailed family information, then check whether the protein has a known structure in the PDB that would allow SCOP or CATH classification. Cross-referencing is particularly important for interpreting molecular docking results, where the structural context of binding sites and domain interfaces matters.

The [domain co-occurrence study](https://pubmed.ncbi.nlm.nih.gov/29293498) demonstrates the value of combining domain information with sequence comparison. The study used domain co-occurrence patterns to improve pairwise sequence comparisons, increasing the sensitivity of BLAST searches and identifying new domain families. This approach shows how domain annotations from databases like Pfam can be used to enhance other bioinformatics analyses.

### Step 4: Run the Analysis and Record Parameters

For Pfam searches, record the version of the database, the search tool used, and the significance thresholds applied. For InterPro searches, record which member databases were included and the version of each. For SCOP and CATH classifications, record the database version and the specific classification assigned at each level of the hierarchy.

Reproducibility requires careful record keeping. The [nf-core Documentation](https://nf-co.re/docs) emphasizes the importance of standardized workflows and version tracking for reproducible analysis. Researchers should apply the same principles when using domain databases, documenting database versions and search parameters so that results can be reproduced or updated as databases change.

### Step 5: Interpret Results Within Database Limitations

Each database has limitations that affect interpretation. Pfam assignments reflect sequence similarity, not structural evidence. SCOP and CATH classifications require structures and may not be available for all proteins. InterPro results aggregate multiple databases and may include conflicting or redundant signatures. Researchers should interpret results within these constraints and avoid overstating conclusions.

For example, a Pfam hit for a domain family does not prove that the protein has the function associated with that family. The hit indicates sequence similarity to known members of the family, but functional confirmation requires experimental evidence. Similarly, a SCOP or CATH classification does not prove evolutionary relationship, particularly at higher levels of the hierarchy where structural similarity may reflect convergent evolution.

## Options and Tradeoffs: Choosing Between SCOP and CATH

SCOP and CATH are often discussed together because both classify protein structures, but they differ in important ways. SCOP places greater emphasis on evolutionary relationships and uses manual curation to define superfamilies. CATH uses more automated methods and provides more frequent updates. The choice between them depends on the research question and the need for currency.

For evolutionary studies, SCOP superfamilies provide a carefully curated view of probable common ancestry. For studies that require up-to-date classification of newly deposited structures, CATH may be more suitable because of its automated pipeline. Researchers who need both perspectives can consult both databases and compare classifications.

The [MMDB and VAST+ study](https://pubmed.ncbi.nlm.nih.gov/24319143) describes an alternative approach to structural classification based on automated structure comparison. VAST+ computes structural similarities between proteins and domains, providing lists of similar structures ranked by the extent of similarity. This approach complements SCOP and CATH by providing a continuously updated view of structural relationships without the delays inherent in curated classification.

## Options and Tradeoffs: Choosing Between Pfam and InterPro

Pfam and InterPro serve different purposes despite their overlap. Pfam is a single database with detailed family information and profile hidden Markov models. InterPro is an integrated resource that searches multiple databases and provides unified results. For detailed analysis of a specific family, Pfam provides more depth. For broad annotation of many sequences, InterPro provides more convenience.

The choice also depends on the need for standardization. InterPro provides consistent output formats that are useful for large-scale analyses and for comparing results across studies. Pfam provides family-specific information that may be lost in the integrated InterPro view. Researchers running genome-scale annotation pipelines often use InterPro because of its standardized output, then consult Pfam for families of particular interest.

The [cancer mutant protein domain study](https://pubmed.ncbi.nlm.nih.gov/34791106) demonstrates the use of domain predictions for large-scale analysis. The study mapped somatic mutations across cancer types to protein domains, using domain predictions for the whole human proteome. This type of analysis requires consistent domain definitions across many proteins, which integrated resources like InterPro can provide.

## Database Selection Matrix for Specific Analysis Types

| Analysis Type | Primary Database | Secondary Database | Key Consideration |
| --- | --- | --- | --- |
| Novel gene annotation | InterPro | Pfam | InterPro provides broad coverage, Pfam adds family detail |
| Evolutionary relationship inference | SCOP | CATH | SCOP superfamilies reflect curated evolutionary judgment |
| Domain architecture comparison | Pfam | InterPro | Pfam domain assignments feed architecture tools |
| Structure validation | CATH | SCOP | CATH updates frequently with new structures |
| Variant interpretation | InterPro | Pfam | Consistent domain definitions support mutation mapping |
| Docking result interpretation | CATH | SCOP | Structural context for binding sites requires structure-based classification |

## Observations and Measurements: What to Record During Domain Analysis

### Database Versions and Search Parameters

Record the exact version of each database used. Pfam, SCOP, CATH, and InterPro all release updates, and classifications can change between versions. A domain assignment from an older version may not match the current version. Record the search tool and its version, as well as any significance thresholds or filtering parameters applied.

### Domain Boundaries and Scores

For Pfam and InterPro searches, record the domain boundaries reported by the search and the scores or expectation values for each match. Domain boundaries define which parts of the sequence correspond to the domain, and these boundaries can affect downstream analyses such as domain architecture comparisons. Scores provide information about the confidence of the match.

### Structural Classification Identifiers

For SCOP and CATH classifications, record the identifiers at each level of the hierarchy. These identifiers allow the classification to be traced and compared across database versions. Record whether the classification was performed by the database curators or by an automated tool, as this affects confidence.

### Cross-References and Evidence Links

Record cross-references to other databases. Pfam entries link to structural databases, functional databases, and literature. SCOP and CATH entries link to PDB structures and to each other. InterPro entries link to all member databases. These cross-references provide evidence for the classification and allow verification.

## Quality Controls and Verification Steps

### Verify Domain Assignments with Multiple Methods

Do not rely on a single database for critical conclusions. If Pfam assigns a domain, check whether InterPro confirms the assignment and whether the domain has a known structure that can be classified in SCOP or CATH. Conflicting assignments may indicate problems with the sequence, the database, or the interpretation.

### Check Domain Boundaries Against Known Features

Compare reported domain boundaries against known features of the protein, such as experimentally determined functional sites, post-translational modification sites, or structural elements. Domain boundaries that overlap known features in unexpected ways may indicate misannotation.

### Validate Structural Classifications

For structural classifications, verify that the structure used for classification is of sufficient quality. Poorly resolved structures or structures with missing residues may produce unreliable classifications. Check the resolution and completeness of the structure before accepting a SCOP or CATH classification.

### Use Reproducible Workflows

The [Galaxy Training Network](https://training.galaxyproject.org/) provides accessible workflow training that emphasizes reproducibility in bioinformatics analysis. Applying these principles to domain database searches ensures that analyses can be repeated and verified. Document all steps in the workflow, including database versions, search parameters, and filtering criteria.

## Common Failure Patterns in Domain Database Use

### Treating Sequence Similarity as Functional Proof

A common error is to conclude that a protein has a particular function because it contains a domain associated with that function. Domain assignments indicate sequence or structural similarity, not demonstrated function. Functional conclusions require experimental evidence or strong supporting data from multiple sources.

### Ignoring Database Version Differences

Domain classifications change as databases are updated. Comparing domain assignments from different database versions without accounting for version differences can produce misleading results. Always record and report database versions.

### Overlooking Domain Architecture Context

Domains do not act in isolation. The [domain architecture comparison study](https://pubmed.ncbi.nlm.nih.gov/19958515) shows that domain architecture, the combination and order of domains in a protein, provides important information about protein function and evolution. Focusing on individual domains without considering the domain architecture can miss key relationships.

### Assuming SCOP and CATH Are Interchangeable

SCOP and CATH use different classification criteria and may assign the same structure to different superfamilies. Researchers who use one database should not assume that the other database will produce the same classification. Cross-referencing both databases provides a more complete picture.

### Misinterpreting InterPro Results

InterPro results aggregate signatures from multiple databases, and these signatures may conflict. A sequence may match a Pfam family and a PROSITE pattern that suggest different functions. Researchers need to examine the individual signatures and understand the evidence for each.

### Neglecting Domain Co-occurrence Information

The [domain co-occurrence study](https://pubmed.ncbi.nlm.nih.gov/29293498) shows that most protein domains tend to appear with a limited number of other domains on the same protein. Ignoring co-occurrence patterns can reduce the sensitivity of sequence comparisons, particularly for phylogenetically distant species. Incorporating co-occurrence information can improve domain family identification and sequence analysis.

## Limitations and Interpretation Boundaries

### Sequence Databases Cannot Predict Structure

Pfam and InterPro classify domains by sequence similarity. These classifications do not provide direct information about three-dimensional structure. A Pfam family may include members with different structures, particularly for families that have diverged substantially. Structural conclusions require structural data and classification in SCOP or CATH.

### Structural Databases Require Structures

SCOP and CATH classify experimentally determined structures. Proteins without structures cannot be classified in these databases. Predicted structures can be submitted for classification, but the reliability of the classification depends on the reliability of the prediction. Researchers should validate predicted structures before using them for classification.

### Domain Boundaries Are Not Always Clear

The [domain interface study](https://pubmed.ncbi.nlm.nih.gov/10708645) shows that domain interfaces have properties similar to protein-protein interfaces, with hydrophobic residues playing a major role. Defining domain boundaries is not always straightforward, and different databases may define boundaries differently. Domain boundary definitions affect downstream analyses, including domain architecture comparisons.

### Classification Hierarchies Reflect Judgment

The higher levels of classification hierarchies, such as fold in SCOP and topology in CATH, reflect judgments about structural similarity that may not correspond to evolutionary relationships. Structural similarity can arise from convergent evolution, and fold classification does not prove common ancestry.

### Automated Structure Comparison Provides Different Information

The [MMDB and VAST+ study](https://pubmed.ncbi.nlm.nih.gov/24319143) describes automated structure comparison as a complement to curated classification. VAST+ identifies structural similarities across the protein structure archive, but these similarities are ranked by extent of similarity instead of by evolutionary inference. Researchers should understand that automated similarity rankings and curated classifications answer different questions.

## Safety and Regulatory Context for Research Applications

Domain database analysis is a computational method that informs research but does not directly produce clinical or regulatory conclusions. However, domain annotations are increasingly used in applications with regulatory implications, including variant interpretation in clinical genetics and target identification in drug development. The [cancer mutant protein domain study](https://pubmed.ncbi.nlm.nih.gov/34791106) demonstrates the use of domain annotations for mapping mutations in cancer, an application with direct clinical relevance.

Researchers using domain annotations for clinical or regulatory purposes should ensure that their analyses meet applicable standards for evidence quality and documentation. Domain assignments should be verified with multiple methods, database versions should be documented, and conclusions should be supported by experimental evidence where available. The [The Carpentries Lessons](https://carpentries.org/lessons) provide foundational training in data management and reproducible analysis that supports these standards.

## Professional Escalation Criteria

### When to Consult a Structural Bioinformatics Specialist

Consult a specialist when domain assignments are ambiguous or conflicting across databases. A specialist can help interpret conflicting evidence and determine whether additional analyses are needed. Escalate when the research question requires structural interpretation that goes beyond standard database searches, such as detailed analysis of domain interfaces or binding sites.

### When to Consult a Domain Expert for the Protein Family

Consult a domain expert when the protein belongs to a family with complex biology or when functional conclusions depend on domain assignments. A domain expert can provide context about family-specific features that are not captured in database annotations.

### When to Reconsider the Analysis Approach

Reconsider the analysis approach when domain database searches produce no hits for a protein that is expected to contain domains. This may indicate that the protein sequence is problematic, that the databases do not cover the relevant domain family, or that the protein uses an unusual domain architecture. The [domain co-occurrence study](https://pubmed.ncbi.nlm.nih.gov/29293498) shows that standard approaches can miss domains, particularly for phylogenetically distant species, and that alternative methods can improve coverage.

### When to Escalate for Regulatory or Clinical Applications

Escalate to qualified professionals when domain annotations will be used in clinical decision-making, regulatory submissions, or other high-stakes applications. Domain assignments alone do not establish clinical significance. Additional evidence, including experimental validation and expert interpretation, is required before domain annotations can support clinical or regulatory conclusions.

## A Practical Decision Framework: Building a Domain Database Selection Protocol for Your Research Pipeline

The previous sections described what each database does, but researchers still face a recurring problem when they sit down to analyze a new protein or genome: how to translate a general research question into a concrete sequence of database searches that produces defensible results. This section provides a structured decision protocol that moves beyond database descriptions and into operational territory. The protocol treats database selection as a series of explicit checkpoints, each with defined inputs, outputs, and escalation criteria. It also includes a record system designed to make domain analysis reproducible and auditable, which matters when results feed into publications, collaborations, or regulatory submissions.

### The Five Checkpoint Decision Protocol

The protocol below organizes database selection into five checkpoints. Each checkpoint requires a specific decision before moving to the next. The protocol assumes the researcher has already defined the biological question and has either a sequence or a structure as the starting material.

#### Checkpoint 1: Determine the Primary Data Type

The first checkpoint separates sequence-based questions from structure-based questions. This decision determines which databases are even eligible for use.

For sequence inputs, the eligible databases are Pfam and InterPro. Both accept protein sequences and return domain family assignments. The choice between them depends on the scale of the analysis and the need for integrated annotation, which the next checkpoint addresses.

For structure inputs, the eligible databases are SCOP and CATH. Both accept PDB identifiers or structure files and return structural classifications. The choice between them depends on whether the research question prioritizes curated evolutionary inference or automated currency.

For predicted structures from tools like AlphaFold, treat the input as a structure but apply additional validation before classification. Predicted structures carry uncertainty that experimental structures do not, and this uncertainty affects the reliability of any downstream classification. The [MMDB and VAST+ study](https://pubmed.ncbi.nlm.nih.gov/24319143) describes automated structure comparison methods that can complement curated classification for predicted structures, but the quality of the prediction remains the limiting factor.

#### Checkpoint 2: Match the Question to the Database Output

The second checkpoint maps the research question to the type of output the database provides. This prevents the common error of using a database that returns the wrong kind of information for the question being asked.

Questions about conserved sequence regions, family membership, and domain architecture point to Pfam. Pfam returns domain family assignments with boundaries and scores, and these assignments feed directly into domain architecture analysis. The [domain architecture comparison study](https://pubmed.ncbi.nlm.nih.gov/19958515) demonstrates how Pfam domain assignments can be used to compare proteins at the domain architecture level, using weighted measures that account for promiscuous domains.

Questions about integrated functional annotation across multiple signature databases point to InterPro. InterPro returns matches from all member databases in a standardized format, which is useful for genome-scale annotation where consistency matters more than depth.

Questions about structural fold, evolutionary relationships, and domain interfaces point to SCOP or CATH. Both return hierarchical classifications, but the interpretation differs. SCOP superfamilies reflect curated evolutionary judgment, while CATH homologous superfamilies reflect a mix of automated and manual methods. The [domain interface study](https://pubmed.ncbi.nlm.nih.gov/10708645) used CATH classifications to analyze the physical and chemical properties of domain-domain interactions, demonstrating the utility of CATH for studying how domains interact within proteins.

Questions about variant interpretation and mutation mapping point to InterPro or Pfam, depending on the need for consistency. The [cancer mutant protein domain study](https://pubmed.ncbi.nlm.nih.gov/34791106) mapped somatic mutations across cancer types to protein domains, using domain predictions for the whole human proteome. This type of analysis requires consistent domain definitions across many proteins, which integrated resources like InterPro can provide.

#### Checkpoint 3: Assess the Need for Cross-Referencing

The third checkpoint determines whether the research question requires information from multiple databases. Many questions do, and the protocol should account for this from the start instead of treating cross-referencing as an afterthought.

A researcher studying a novel protein might start with an InterPro search to get broad functional annotation, then consult Pfam for detailed family information, then check whether the protein has a known structure in the PDB that would allow SCOP or CATH classification. This sequence of searches produces a more complete picture than any single database can provide.

The [domain co-occurrence study](https://pubmed.ncbi.nlm.nih.gov/29293498) demonstrates the value of combining domain information with sequence comparison. The study used domain co-occurrence patterns to improve pairwise sequence comparisons, increasing the sensitivity of BLAST searches and identifying new domain families. This approach shows how domain annotations from databases like Pfam can be used to enhance other bioinformatics analyses, and it illustrates why cross-referencing should be planned instead of improvised.

#### Checkpoint 4: Run the Searches and Record Everything

The fourth checkpoint is execution with documentation. The protocol requires recording the database version, the search tool and its version, the search parameters including significance thresholds, and the date of the search. This information is essential for reproducibility and for interpreting results when databases change.

For Pfam searches, record the database version, the search tool used, and the expectation value or score thresholds applied. For InterPro searches, record which member databases were included and the version of each. For SCOP and CATH classifications, record the database version and the specific classification assigned at each level of the hierarchy.

The [nf-core Documentation](https://nf-co.re/docs) emphasizes the importance of standardized workflows and version tracking for reproducible analysis. The same principles apply to domain database searches. Document all steps in the workflow, including database versions, search parameters, and filtering criteria, so that the analysis can be repeated or updated as databases change.

#### Checkpoint 5: Interpret Within Database Limitations

The fifth checkpoint is interpretation with explicit acknowledgment of what each database can and cannot support. This prevents overstatement and keeps conclusions within the bounds of the evidence.

A Pfam hit indicates sequence similarity to known members of a domain family. It does not prove that the protein has the function associated with that family. Functional conclusions require experimental evidence or strong supporting data from multiple sources.

A SCOP or CATH classification indicates structural similarity according to the criteria of that database. It does not prove evolutionary relationship, particularly at higher levels of the hierarchy where structural similarity may reflect convergent evolution.

An InterPro result aggregates signatures from multiple databases, and these signatures may conflict. A sequence may match a Pfam family and a PROSITE pattern that suggest different functions. Researchers need to examine the individual signatures and understand the evidence for each.

### A Record System for Domain Database Analysis

Reproducible domain analysis requires more than remembering which database was used. The record system below provides a structured format for documenting each analysis. This format supports both personal reproducibility and external audit, which matters when results are shared with collaborators or included in publications.

#### The Domain Analysis Record

For each protein or structure analyzed, create a record with the following fields:

| Field | Content | Example |
| --- | --- | --- |
| Protein identifier | The sequence or structure identifier used | UniProt P12345 or PDB 1ABC |
| Database | The database searched | Pfam 35.0 |
| Search tool | The tool and version used | HMMER 3.3.2 |
| Search date | The date the search was run | 2024-03-15 |
| Significance threshold | The threshold applied | E-value less than 1e-5 |
| Domain assignments | The domains identified with boundaries and scores | PF00001, residues 45-210, score 150.2 |
| Cross-references | Links to other databases | InterPro IPR000001, SCOP d1abca_ |
| Interpretation notes | Context about how the assignment was interpreted | Assignment consistent with known function from literature |

This record format captures the information needed to reproduce the analysis and to interpret the results in context. It also provides a basis for comparing results across database versions, which is important because domain classifications change as databases are updated.

#### Version Tracking and Update Protocol

Databases release updates on different schedules, and classifications can change between versions. A domain assignment from an older version may not match the current version. The record system should include a protocol for tracking versions and for updating analyses when databases change.

For ongoing projects, schedule periodic re-runs of domain searches when major database versions are released. Compare the new results with the previous results and document any changes in domain assignments. Changes may reflect improved models, new family definitions, or corrections to previous classifications.

The [Galaxy Training Network](https://training.galaxyproject.org/) provides accessible workflow training that emphasizes reproducibility in bioinformatics analysis. Applying these principles to domain database searches ensures that analyses can be repeated and verified. Document all steps in the workflow, including database versions, search parameters, and filtering criteria.

### Troubleshooting Common Protocol Failures

Even with a structured protocol, analyses can fail in predictable ways. The troubleshooting guidance below addresses the most common failure patterns and provides concrete steps for diagnosis and correction.

#### Failure Pattern: No Domain Hits for an Expected Domain-Containing Protein

When a protein expected to contain domains returns no hits, the first step is to check the sequence itself. The sequence may contain errors, may be truncated, or may be a fragment that does not include the expected domain region. Check the sequence against the source database to verify its integrity.

If the sequence is intact, the next step is to check database coverage. The [domain co-occurrence study](https://pubmed.ncbi.nlm.nih.gov/29293498) shows that standard approaches can miss domains, particularly for phylogenetically distant species. The study found that incorporating domain co-occurrence information increased the number of significant BLAST hits by 14 percent and increased the proteome area covered with a domain by 25 percent. If standard searches return no hits, consider using domain co-occurrence information or alternative search methods.

The third step is to check whether the protein uses an unusual domain architecture. The [domain architecture comparison study](https://pubmed.ncbi.nlm.nih.gov/19958515) shows that promiscuous domains, which appear in many different protein contexts, can complicate domain identification. If the protein contains a promiscuous domain, standard searches may miss it because the domain is not strongly associated with any particular protein context.

#### Failure Pattern: Conflicting Domain Assignments Across Databases

Conflicting assignments are common and do not necessarily indicate an error. Different databases use different classification criteria and may define domains differently. The first step is to document the conflict and examine the evidence for each assignment.

Check whether the conflicting assignments come from databases with different classification philosophies. Pfam classifies by sequence similarity, while SCOP and CATH classify by structure. A domain that is clearly defined by structure may not have a clear sequence signature, and vice versa. The [domain interface study](https://pubmed.ncbi.nlm.nih.gov/10708645) shows that domain boundaries are not always clear and that different databases may define boundaries differently.

If the conflict persists after examining the evidence, consult a specialist. A structural bioinformatics specialist can help interpret conflicting evidence and determine whether additional analyses are needed.

#### Failure Pattern: Domain Boundaries That Do Not Match Known Features

When reported domain boundaries overlap known features of the protein in unexpected ways, the boundaries may be misannotated. Compare reported boundaries against experimentally determined functional sites, post-translational modification sites, or structural elements.

The [domain interface study](https://pubmed.ncbi.nlm.nih.gov/10708645) shows that domain interfaces have characteristic properties, including hydrophobic residues and positively charged arginine residues. These properties can help identify domain boundaries when database assignments are uncertain. If reported boundaries do not align with known features, consider whether the database assignment is reliable or whether alternative boundary definitions are more appropriate.

#### Failure Pattern: Results That Cannot Be Reproduced

When a colleague or reviewer cannot reproduce the results of a domain analysis, the most common cause is incomplete documentation. The record system described above addresses this by requiring database versions, search parameters, and search dates for every analysis.

If results cannot be reproduced despite complete documentation, check whether the database has been updated since the original search. Domain classifications change as databases are updated, and a search run against an older version may produce different results than a search run against the current version. Re-run the search against the current version and compare the results.

### Integrating the Protocol with Existing Workflows

The decision protocol is designed to integrate with existing bioinformatics workflows instead of replace them. Researchers already using pipelines for sequence analysis, structure prediction, or genome annotation can add the protocol as a structured decision layer that determines which domain databases to query and how to document the results.

The [Bioconductor](https://bioconductor.org/) project provides official packages and workflows for genomic analysis, and the [nf-core Documentation](https://nf-co.re/docs) describes community pipeline standards for reproducible analysis. Both resources support the integration of domain database searches into larger analysis pipelines. The protocol described here provides the decision framework that determines which database searches to include in these pipelines.

The [The Carpentries Lessons](https://carpentries.org/lessons) provide foundational training in data management and reproducible analysis that supports the record-keeping requirements of the protocol. Researchers who have completed this training will find the record system straightforward to implement.

### When to Escalate Beyond the Protocol

The protocol covers standard domain analysis scenarios, but some situations require escalation beyond what the protocol provides. The escalation criteria below define when to seek additional expertise.

Escalate to a structural bioinformatics specialist when domain assignments are ambiguous or conflicting across databases and the conflict affects the research conclusion. A specialist can help interpret conflicting evidence and determine whether additional analyses are needed.

Escalate to a domain expert for the protein family when the protein belongs to a family with complex biology or when functional conclusions depend on domain assignments. A domain expert can provide context about family-specific features that are not captured in database annotations.

Escalate to qualified professionals when domain annotations will be used in clinical decision-making, regulatory submissions, or other high-stakes applications. Domain assignments alone do not establish clinical significance. Additional evidence, including experimental validation and expert interpretation, is required before domain annotations can support clinical or regulatory conclusions. The [cancer mutant protein domain study](https://pubmed.ncbi.nlm.nih.gov/34791106) demonstrates the use of domain annotations for mapping mutations in cancer, an application with direct clinical relevance, and it illustrates the need for careful interpretation in this context.

### Practical Implementation Steps

Implementing the protocol requires a small investment in setup and documentation. The steps below provide a practical path to implementation.

First, create a template for the Domain Analysis Record described above. This template can be a spreadsheet, a text file, or a database table, depending on the scale of the analysis. The template should include all fields from the record system.

Second, define the significance thresholds and search parameters for each database. These thresholds should be consistent across analyses to support comparison. Record the thresholds in the analysis record.

Third, establish a version tracking protocol. Decide how often to re-run domain searches and how to document changes in database versions. The protocol should include a process for comparing results across versions and for updating conclusions when classifications change.

Fourth, integrate the protocol into existing workflows. Identify the points in current analysis pipelines where domain database searches occur and add the protocol as a structured decision layer. Document the integration so that collaborators and reviewers understand how database selection decisions were made.

Fifth, test the protocol on a known protein or structure. Run the protocol on a protein with well-characterized domains and verify that the results match the known biology. This validation step confirms that the protocol is working correctly before it is applied to novel proteins.

The [EMBL-EBI Training](https://www.ebi.ac.uk/training) resources provide instruction on using InterPro and other EBI databases for sequence analysis, which can help researchers implement the protocol effectively. The [NCBI Data Resources](https://www.ncbi.nlm.nih.gov/) provide access to sequence databases and search tools that complement domain database analysis, allowing researchers to retrieve sequences and perform initial similarity searches before submitting them to domain databases.

## Frequently Asked Questions

### What is the difference between Pfam and InterPro?

Pfam is a single database of protein domain families built from multiple sequence alignments and profile hidden Markov models. InterPro is an integrated resource that searches multiple signature databases, including Pfam, and provides unified results. Pfam provides detailed family information, while InterPro provides broad coverage and standardized output. Researchers often use InterPro for genome-scale annotation and Pfam for detailed analysis of specific families.

### When should I use SCOP instead of CATH?

Use SCOP when the research question emphasizes evolutionary relationships and when curated classification is important. SCOP superfamilies reflect expert judgment about probable common ancestry. Use CATH when the research question requires up-to-date classification of newly deposited structures or when automated classification is preferred. CATH updates more frequently because of its automated pipeline. For many questions, consulting both databases provides the most complete picture.

### Can I classify a predicted structure in SCOP or CATH?

Predicted structures can be submitted for classification in SCOP or CATH, but the reliability of the classification depends on the reliability of the prediction. Poorly predicted structures may produce unreliable classifications. Validate predicted structures using appropriate quality metrics before submitting them for classification. The [MMDB and VAST+ study](https://pubmed.ncbi.nlm.nih.gov/24319143) describes automated structure comparison methods that can complement curated classification for predicted structures.

### How do I choose domain boundaries for downstream analysis?

Domain boundaries reported by Pfam or InterPro are a reasonable starting point, but they may not match boundaries defined by structural classification in SCOP or CATH. For analyses that depend on precise domain boundaries, such as domain architecture comparisons, verify boundaries against structural data when available. The [domain interface study](https://pubmed.ncbi.nlm.nih.gov/10708645) shows that domain interfaces have characteristic properties that can help identify domain boundaries.

### What does a Pfam hit tell me about protein function?

A Pfam hit indicates that the protein sequence is similar to known members of a domain family. This provides evidence about possible function but does not prove function. Functional conclusions require experimental evidence or strong supporting data from multiple sources. The hit also provides information about conserved sequence regions that may be functionally important.

### How do I handle conflicting domain assignments from different databases?

Conflicting assignments are common and do not necessarily indicate an error. Different databases use different classification criteria and may define domains differently. Examine the evidence for each assignment, including sequence similarity, structural data, and functional annotations. If conflicts persist, consult a specialist or use additional analysis methods such as the domain architecture comparison approach described in the [domain architecture study](https://pubmed.ncbi.nlm.nih.gov/19958515).

### What is domain co-occurrence and why does it matter?

Domain co-occurrence refers to the tendency of certain domains to appear together in the same protein. The [domain co-occurrence study](https://pubmed.ncbi.nlm.nih.gov/29293498) shows that most protein domains appear with a limited number of other domains. This feature can be used to improve sequence comparisons and to identify new domain families. Domain co-occurrence also provides information about protein function and evolution.

### How do I document domain database searches for reproducibility?

Record the database version, the search tool and its version, the search parameters including significance thresholds, and the date of the search. Record all results, including domain boundaries and scores. Store the records with the analysis data so that the analysis can be reproduced or updated. The [nf-core Documentation](https://nf-co.re/docs) and the [Galaxy Training Network](https://training.galaxyproject.org/) provide guidance on reproducible analysis workflows.

## Related Bioinformatics Guides

- [Genomic Prediction in Livestock: A Decision Framework for Breeders](/knowledge/bioinformatics/genomic-prediction-in-livestock-a-decision-framework-for-breeders)
- [RNA-Seq Databases: Accessing and Using Public RNA-Seq Data](/knowledge/bioinformatics/rna-seq-databases-accessing-and-using-public-rna-seq-data)
- [Selecting Persistent Identifiers for Research Data: A Decision Framework](/knowledge/bioinformatics/selecting-persistent-identifiers-for-research-data-a-decision-framework)
- [Cell Cycle Checkpoints: A Decision Framework for Identifying Phase-Specific Defects](/knowledge/bioinformatics/cell-cycle-checkpoints-a-decision-framework-for-identifying-phase-specific-defects)
- [Single-Cell Sequencing Databases: Resources for Data Sharing and Exploration](/knowledge/bioinformatics/single-cell-sequencing-databases-resources-for-data-sharing-and-exploration)

## 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.
- [MMDB and VAST+: tracking structural similarities between macromolecular complexes.](https://pubmed.ncbi.nlm.nih.gov/24319143). Nucleic acids research, 2014.
- [Protein domain interfaces: characterization and comparison with oligomeric protein interfaces.](https://pubmed.ncbi.nlm.nih.gov/10708645). Protein engineering, 2000.
- [Improving pairwise comparison of protein sequences with domain co-occurrence.](https://pubmed.ncbi.nlm.nih.gov/29293498). PLoS computational biology, 2018.
- [Protein comparison at the domain architecture level.](https://pubmed.ncbi.nlm.nih.gov/19958515). BMC bioinformatics, 2009.
- [DCMP: database of cancer mutant protein domains.](https://pubmed.ncbi.nlm.nih.gov/34791106). Database : the journal of biological databases and curation, 2021.

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