# How to Build a Custom Protein Sequence Database for Proteomics: From Raw Genomes to FASTA Files

Researchers working with non-model organisms, variant-containing samples, or immunoglobulin repertoires face a common problem: standard reference proteomes do not contain every protein sequence that may be present in their samples. A mass spectrometry search is only as good as the database it searches against. If a peptide sequence is absent from the FASTA file, it cannot be identified, regardless of instrument quality or search algorithm performance. Building a custom protein sequence database from genomic or transcriptomic data solves this problem by generating a search space tailored to the specific organism, condition, or biological question under investigation.

This article provides a standardized protocol for constructing custom FASTA databases. It covers data acquisition from public repositories, sequence extraction from genomes and transcriptomes, six-frame translation, variant incorporation, contaminant addition, decoy generation, quality control, and reproducibility practices. The workflow is designed for biology students, researchers, laboratory professionals, and life-science practitioners who need a reliable template instead of ad hoc scripts.

## Scope and Reader Context

The target reader is a researcher who has mass spectrometry data or plans to generate it, and who needs a protein sequence database that reflects the actual proteome of their sample. This situation arises in several common scenarios. First, working with a non-model organism whose genome is sequenced but whose proteome is not curated in a major database. Second, studying a model organism but needing to include sequence variants, splice isoforms, or somatic mutations that are absent from the canonical reference proteome. Third, analyzing metaproteomics samples where multiple organisms contribute proteins and the database must balance comprehensiveness against size. Fourth, studying highly divergent sequences such as antibody repertoires where standard databases are known to miss identifications.

The protocol described here assumes basic familiarity with command-line computing, FASTA format, and the general principles of mass spectrometry-based proteomics. Readers who need to strengthen their foundational bioinformatics skills can consult training materials from [The Carpentries](https://carpentries.org/lessons) for shell and programming fundamentals, or [EMBL-EBI Training](https://www.ebi.ac.uk/training) for data-resource specific instruction. The [Galaxy Training Network](https://training.galaxyproject.org/) offers accessible workflow tutorials that can serve as a starting point for researchers who prefer graphical interfaces over command-line tools.

## Why Custom Databases Matter in Proteomics

The central role of the protein sequence database in peptide identification cannot be overstated. During a typical proteomics search, experimental tandem mass spectra are compared against theoretical spectra generated from every peptide in the database. Peptides that are not represented in the database produce no theoretical spectra and therefore cannot be matched. This fundamental limitation means that database construction directly determines the upper bound of identification performance.

A review of protein sequence database construction for metaproteomics emphasizes that building a reference database that is both comprehensive and not overly large is critical for the peptide identification step. The availability of well-curated reference databases and tools for custom database construction is essential to enhance the performance of metaproteomics analyses. This principle extends beyond metaproteomics to any experiment where the sample proteome diverges from the reference proteome.

The consequences of an incomplete database are not limited to missed identifications. When a sample contains proteins that are absent from the database, the search engine may assign those spectra to the closest available sequence, producing false positive identifications that are statistically indistinguishable from true matches. This problem is particularly acute for highly divergent sequences. A study of the antibody repertoire in colorectal cancer demonstrated that standard protein databases do not include the highly divergent sequences of somatic rearranged immunoglobulin genes, and this absence can lead to misidentifications in a mass spectrometry search. The authors developed a customized antibody database construction method using RNA-seq reads aligned to immunoglobulin genes, which identified 1,940 distinct antibody peptides from 90 tumor samples.

Custom database construction is therefore not an optional refinement but a necessary step for experiments involving non-model organisms, variant-rich samples, immunoglobulin repertoires, or metaproteomics communities.

## Core Principles of Database Construction

### Comprehensiveness Versus Size Tradeoff

The fundamental tension in database construction is between sensitivity and specificity. A larger database increases the chance that any given peptide will be present, which improves sensitivity. However, a larger database also increases the search space, which raises the number of candidate peptides for each spectrum and increases the likelihood of random matches. This tradeoff is managed through statistical corrections such as the target-decoy approach, but the underlying principle remains: the database should contain all relevant protein sequences and as little irrelevant sequence as possible.

For metaproteomics, the review literature specifically notes that constructing a reference protein sequence database that is both comprehensive and not overly large is critical for the peptide identification step. This balance requires careful decisions about which organisms to include, which protein isoforms to retain, and how to handle redundant sequences.

### Sequence Source Hierarchy

Protein sequences can be derived from several sources, each with distinct advantages and limitations. The choice of source depends on data availability and the biological question.

Curated reference proteomes from [NCBI](https://www.ncbi.nlm.nih.gov/) provide the highest quality annotations for well-studied organisms. These databases are manually reviewed and contain functional annotations, but they are incomplete for non-model organisms and do not include sample-specific variants.

Genome-based prediction uses gene prediction algorithms to identify coding sequences in a genome assembly. This approach captures the full set of predicted proteins for any organism with a sequenced genome, but prediction accuracy varies and pseudogenes or misannotated genes may introduce errors.

Transcriptome-based construction uses RNA-seq data to assemble expressed transcripts, which are then translated to protein sequences. This approach captures sample-specific expression and splice isoforms, but requires matching transcript and proteomics data from the same or closely related samples.

Six-frame translation of genomic sequence captures all possible open reading frames without relying on gene prediction. This approach is comprehensive but generates enormous databases with many non-coding sequences, making it a last resort for organisms with no annotation.

Proteogenomic approaches combine genomic or transcriptomic evidence with mass spectrometry data to refine protein databases. The antibody repertoire study exemplifies this approach by using RNA-seq reads to construct a graph structure that captures the high diversity of immunoglobulin genes in a compact manner.

### Contaminant Sequences

Every proteomics experiment contains contaminant proteins from handling, reagents, and the laboratory environment. Common contaminants include keratins from skin and hair, serum proteins from cell culture media, and trypsin or other proteases used for digestion. These contaminants appear in every sample regardless of the organism under study.

Adding a contaminant database to the search database serves two purposes. First, it allows contaminant peptides to be identified and excluded from downstream analysis. Second, it prevents contaminant spectra from being incorrectly assigned to sample proteins. Most search engines support the annotation of contaminant proteins so that they can be flagged in the output.

The contaminant database should be appended to the sample-specific database before the addition of decoy sequences. This ordering ensures that the decoy generation process treats contaminants and sample proteins identically.

### Decoy Sequences

Decoy sequences are reversed or shuffled versions of target sequences that are used to estimate the false discovery rate. The target-decoy approach works by searching spectra against a combined database containing both target and decoy sequences. Matches to decoy sequences are assumed to be false positives, and the ratio of decoy matches to target matches provides an estimate of the false discovery rate at any given score threshold.

Decoy generation must be performed after all target sequences have been assembled, including contaminants. The decoy generation method should preserve the amino acid composition and length distribution of the target sequences. Reversal is generally preferred over shuffling because it preserves the distribution of amino acids at each position and avoids the creation of peptides that are identical to target peptides.

## Data Acquisition from Public Repositories

### NCBI Resources

The [National Center for Biotechnology Information](https://www.ncbi.nlm.nih.gov/) provides the primary public repositories for genomic sequence data, including GenBank for annotated sequences, the Sequence Read Archive for raw sequencing data, and the Reference Sequence collection for curated assemblies. NCBI also hosts the BLAST search system for sequence similarity searching and the Genome Browser for visualizing genomic features.

For custom database construction, NCBI serves as the primary source for genome assemblies, annotated coding sequences, and transcript sequences. The RefSeq collection provides curated, non-redundant sequence sets for many organisms, which can serve as a starting point for database construction. For non-model organisms, the Whole Genome Shotgun assembly database contains genome assemblies that may not yet have comprehensive annotation.

The choice between GenBank and RefSeq sequences matters for database construction. RefSeq sequences are curated and non-redundant, making them suitable for direct use. GenBank sequences may contain redundant entries and require additional filtering. For organisms without RefSeq representation, GenBank is the only option.

### EMBL-EBI Resources

The [European Bioinformatics Institute](https://www.ebi.ac.uk/training) provides complementary resources to NCBI, including the European Nucleotide Archive for sequence data and UniProt for protein sequences. EMBL-EBI training materials cover the use of these resources for sequence retrieval and analysis.

UniProt is particularly valuable for database construction because it provides reviewed protein sequences with functional annotations. The UniProtKB/Swiss-Prot section contains manually reviewed entries, while the TrEMBL section contains automatically annotated entries. For well-studied organisms, the reviewed entries provide a high-quality starting point. For non-model organisms, the unreviewed entries may be the only available protein sequences.

### Downloading Genomes and Annotations

Genome assemblies are typically downloaded as FASTA files containing chromosome or scaffold sequences. Annotations are downloaded as GFF or GTF files that describe the locations of genes, transcripts, and coding sequences on the genome. The combination of genome sequence and annotation file allows extraction of protein sequences using tools such as gffread or custom scripts.

When downloading from NCBI, the assembly accession should be recorded along with the download date. Genome assemblies are frequently updated, and reproducibility requires that the exact assembly version be documented. The same principle applies to annotation files, which may be updated independently of the genome assembly.

For transcriptome-based construction, RNA-seq reads are downloaded from the Sequence Read Archive. The read files can be large, and download times should be factored into project planning. The choice of read files depends on the analysis plan: paired-end reads are generally preferred for transcript assembly, and the sequencing depth affects the completeness of the resulting transcriptome.

## Step-by-Step Workflow for Database Construction

### Step 1: Define the Biological Scope

Before downloading any data, define the biological scope of the database. This definition should address four questions. First, what organism or organisms are expected in the sample? For metaproteomics, this may be a community of organisms instead of a single species. Second, what sequence variants should be included? This includes splice isoforms, single amino acid variants, somatic mutations, and immunoglobulin rearrangements. Third, what is the expected dynamic range of protein abundances? This affects whether rare isoforms need to be included. Fourth, what is the acceptable database size? This is constrained by computational resources and search time.

The biological scope should be documented in the laboratory notebook or project management system before data acquisition begins. This documentation serves as the basis for later quality control and troubleshooting.

### Step 2: Acquire Genomic or Transcriptomic Data

For genome-based construction, download the genome assembly and annotation file from [NCBI](https://www.ncbi.nlm.nih.gov/) or EMBL-EBI. Verify the integrity of downloaded files using checksums provided by the repository. Record the assembly accession, annotation version, and download date.

For transcriptome-based construction, download RNA-seq reads from the Sequence Read Archive. If the RNA-seq data were generated in the same laboratory, verify that the sample metadata matches the proteomics sample metadata. Mismatched samples are a common source of error in proteogenomic studies.

For organisms without existing genomic resources, de novo genome assembly or transcriptome assembly may be required. This is a substantial undertaking that should be planned separately from database construction. The [Galaxy Training Network](https://training.galaxyproject.org/) provides tutorials for genome assembly and annotation that can guide this process.

### Step 3: Extract Coding Sequences

If an annotation file is available, extract the coding sequences using the annotation coordinates. The extraction process should produce one FASTA entry per transcript, with the header containing the gene identifier, transcript identifier, and any available functional annotation.

For genomes without annotation, gene prediction tools can identify putative coding sequences. The accuracy of gene prediction varies by organism and tool, and predicted genes should be treated with caution. A common approach is to combine gene prediction with evidence from RNA-seq data or homology to related organisms.

For transcriptome-based construction, assemble the RNA-seq reads into transcripts using a transcript assembler. The assembled transcripts are then translated to protein sequences using the appropriate genetic code. The translation frame is determined by identifying the longest open reading frame in each transcript.

### Step 4: Translate to Protein Sequences

Nucleotide coding sequences must be translated to protein sequences using the appropriate genetic code. The standard genetic code is used for most organisms, but alternative codes are used for mitochondria, chloroplasts, and some microbial lineages. Using the wrong genetic code produces incorrect protein sequences and leads to missed identifications.

Translation tools typically require the nucleotide sequence and the genetic code as input. The output is a protein FASTA file with headers that preserve the transcript or gene identifiers. The translation frame should be recorded for each sequence, particularly when six-frame translation is used.

For six-frame translation, the genome sequence is translated in all six reading frames. This approach generates six protein sequences for every genomic region, three from the forward strand and three from the reverse strand. The resulting database is large and contains many sequences that are not translated in vivo, but it ensures that no potential open reading frame is missed.

### Step 5: Add Sample-Specific Variants

Sample-specific variants can be incorporated into the database using several approaches. The choice of approach depends on the type of variant and the available data.

Single amino acid variants identified from DNA or RNA sequencing can be incorporated by modifying the reference protein sequences. Each variant position is changed to the variant amino acid, and the resulting sequence is added to the database as a separate entry. The header should indicate the variant position and amino acid change.

Splice isoforms can be incorporated by translating all observed transcript isoforms instead of only the canonical transcript. RNA-seq data can identify which isoforms are expressed in the sample, allowing the database to focus on expressed isoforms.

Immunoglobulin and T-cell receptor sequences require specialized approaches because of their extreme diversity. The antibody repertoire study used a graph-based approach to capture the diversity of immunoglobulin genes in a compact manner. The graph structure was converted to a format suitable for searching with mass spectrometry search tools. This approach is more complex than simple variant addition and requires specialized software.

### Step 6: Append Contaminant Sequences

Append the contaminant protein sequences to the sample-specific database. The contaminant database should include common laboratory contaminants such as keratins, serum proteins, and digestion enzymes. Many proteomics software packages provide a standard contaminant database that can be used directly.

The contaminant entries should be clearly marked in the FASTA headers so that they can be identified in the search results. Common conventions include prefixing contaminant headers with "CON" or "contam" or using a specific annotation that the search engine recognizes.

### Step 7: Generate Decoy Sequences

Generate decoy sequences from the combined target database that includes both sample-specific and contaminant sequences. The decoy generation method should be consistent with the search engine requirements. Most search engines support reversed or shuffled decoys, and the choice should be documented.

The decoy sequences should be appended to the target sequences in the final database file. The combined target-decoy database is then used for the mass spectrometry search. The search engine uses the decoy matches to estimate the false discovery rate.

### Step 8: Validate the Database

Validation is a critical step that is often skipped. The validation process should verify that the database is in the correct format, contains the expected number of sequences, and produces sensible results in a test search.

Format validation checks that the FASTA file is properly formatted, with headers starting with ">" and sequences containing only valid amino acid characters. The file should not contain empty sequences or duplicate headers.

Content validation checks that the database contains the expected sequences. This can be done by counting sequences, checking for the presence of specific proteins, and verifying that the sequence length distribution matches expectations.

Search validation runs a small test search against the database using known spectra. The test search should identify expected proteins and produce reasonable false discovery rates. If the test search fails, the database construction process should be reviewed.

## At a Glance

| Decision Point | Recommended Approach | Key Consideration |
|---|---|---|
| Sequence source | Curated proteome if available, otherwise genome-based prediction or transcriptome assembly | Curated sources are highest quality but incomplete for non-model organisms |
| Variant inclusion | Add sample-specific variants as separate FASTA entries | Variants from DNA or RNA sequencing must be matched to the correct reference sequence |
| Contaminant addition | Append standard contaminant database before decoy generation | Contaminants appear in every sample and must be identifiable in search results |
| Decoy generation | Use reversed sequences after all targets are assembled | Decoy method must match search engine requirements |
| Database size | Balance comprehensiveness against search space | Larger databases increase sensitivity but also increase false discovery risk |
| Documentation | Record all source accessions, versions, and parameters | Reproducibility requires complete provenance tracking |

## Tools and Platforms for Database Construction

### Command-Line Tools

Command-line tools provide the most flexible approach to database construction. The core tools include sequence retrieval utilities, sequence manipulation tools, and custom scripts for variant incorporation.

The NCBI command-line tools provide access to Entrez databases for sequence retrieval. These tools can download genome assemblies, protein sequences, and annotation files programmatically. The [NCBI](https://www.ncbi.nlm.nih.gov/) website also provides a web interface for manual downloads.

Sequence manipulation tools such as the EMBOSS package provide utilities for translation, reverse complementation, and sequence filtering. These tools are well documented and can be combined into pipelines using shell scripts.

Custom scripts are often needed for variant incorporation and database assembly. Python and R are the most common languages for these scripts. The [Bioconductor](https://bioconductor.org/) project provides R packages for genomic analysis that can be used for database construction workflows.

### Workflow Managers

Workflow managers provide reproducibility and scalability for database construction pipelines. The [nf-core](https://nf-co.re/docs) community provides standardized pipelines for bioinformatics analysis, including proteomics workflows. These pipelines follow community standards for configuration, usage, and reproducibility.

The [Galaxy Training Network](https://training.galaxyproject.org/) provides accessible workflow tutorials that can be used to learn database construction without command-line experience. Galaxy workflows can be saved and shared, providing a reproducible record of the analysis.

### Containerization

Containerization ensures that database construction tools run consistently across different computing environments. Containers package the software and its dependencies into a single image that can be run on any system with the container runtime installed.

The use of containers is particularly important for reproducibility because software versions and dependencies change over time. A database constructed with one version of a tool may differ from a database constructed with a different version. Containerization freezes the software environment and ensures that the same tool versions are used every time.

## Quality Control and Validation

### Sequence-Level Checks

Sequence-level quality control verifies that individual sequences in the database are correct. The checks include verifying that sequences contain only valid amino acid characters, that sequences have reasonable lengths, and that sequences do not contain internal stop codons.

Internal stop codons indicate that the translation frame is incorrect or that the sequence contains a sequencing error. Sequences with internal stop codons should be reviewed and either corrected or removed from the database.

The sequence length distribution should be examined for anomalies. A database that contains many very short sequences may indicate a problem with translation or sequence extraction. A database that contains unusually long sequences may indicate that multiple proteins were concatenated.

### Database-Level Checks

Database-level quality control verifies that the database as a whole is appropriate for the intended search. The checks include verifying the total number of sequences, the number of target and decoy sequences, and the presence of contaminant sequences.

The target and decoy sequence counts should be equal if the decoy generation process is working correctly. An imbalance indicates a problem with the decoy generation process.

The database size should be compared against expectations. A database that is much larger than expected may contain redundant sequences or spurious translations. A database that is much smaller than expected may be missing important sequences.

### Search-Based Validation

Search-based validation runs a test search against the database to verify that it produces sensible results. The test search should use spectra from a well-characterized sample or a subset of the experimental spectra.

The search results should be examined for expected proteins, reasonable identification rates, and appropriate false discovery rates. If the search produces unexpected results, the database construction process should be reviewed.

The [EMBL-EBI Training](https://www.ebi.ac.uk/training) resources provide guidance on interpreting mass spectrometry search results and validating database searches.

## Records and Documentation

### Provenance Tracking

Provenance tracking records the origin of every sequence in the database. This information is essential for reproducibility and for troubleshooting identification issues.

The provenance record should include the source database and accession for each sequence, the version of the source database, the download date, and the tool versions used for processing. This information can be stored in a spreadsheet or in the FASTA headers themselves.

FASTA headers should be structured to include the source identifier, any processing steps applied, and the date of construction. A consistent header format simplifies downstream analysis and troubleshooting.

### Version Control

Version control is essential for database construction because the process is iterative. The database will be modified as new data become available or as errors are discovered.

Version control systems such as Git can track changes to the database construction scripts and documentation. The database files themselves are typically too large for version control, but the scripts and parameters used to generate them should be version controlled.

Each version of the database should be assigned a version number and documented with the date, the source data versions, and the changes from the previous version.

### Laboratory Notebook Documentation

The laboratory notebook should contain a complete record of the database construction process. This record should include the biological scope, the data sources, the processing steps, the quality control results, and the final database version.

The notebook record serves as the primary documentation for the database and is essential for reproducing the analysis or for responding to reviewer questions.

## Common Failure Patterns

### Incomplete Sequence Sources

The most common failure pattern is using an incomplete sequence source. This occurs when the database is built from a curated proteome that does not contain all proteins expressed in the sample. The result is missed identifications and potentially false identifications.

This failure is particularly common for non-model organisms, where the curated proteome may contain only a fraction of the expressed proteins. The solution is to supplement the curated proteome with genome-based predictions or transcriptome-based sequences.

### Incorrect Genetic Code

Using the wrong genetic code for translation produces incorrect protein sequences. This failure is common for mitochondrial proteins, which use a different genetic code than nuclear proteins.

The solution is to verify the genetic code for each sequence source and to use the appropriate code for translation. The genetic code should be documented in the provenance record.

### Contaminant Sequences Missing

Omitting contaminant sequences from the database leads to contaminant spectra being assigned to sample proteins. This produces false identifications that are difficult to detect.

The solution is to always include a standard contaminant database in the search database. The contaminant database should be appended before decoy generation.

### Decoy Generation Errors

Errors in decoy generation can produce decoy sequences that are not truly reversed or shuffled, leading to incorrect false discovery rate estimates. This failure is often caused by using an inappropriate decoy generation method or by generating decoys from a subset of the target sequences.

The solution is to use a validated decoy generation method and to verify that the decoy sequences are properly generated.

### Redundant Sequences

Redundant sequences in the database inflate the database size and can skew false discovery rate estimates. This failure is common when combining multiple sequence sources without removing duplicates.

The solution is to remove redundant sequences during database construction. Sequence clustering tools can identify and remove sequences that are identical or nearly identical.

### Header Parsing Failures

Inconsistent FASTA headers can cause failures in downstream analysis tools that parse the headers. This failure is common when combining sequences from multiple sources with different header formats.

The solution is to standardize the header format during database construction. The header should contain the source identifier, the sequence type, and any relevant annotations in a consistent format.

## Limitations and Interpretation

### Database Size Limits

The database size is limited by computational resources and search time. A database that is too large will make the search computationally infeasible or will produce unacceptably high false discovery rates.

The review of metaproteomics database construction notes that the database must be comprehensive but not overly large. This balance requires careful selection of sequences and removal of redundant or irrelevant entries.

### Annotation Quality

The quality of the database depends on the quality of the underlying annotation. Gene prediction errors, misannotated transcripts, and incomplete assemblies all propagate into the protein database.

For non-model organisms, the annotation quality may be low, and the database may contain errors. These errors can produce false identifications or missed identifications. The limitations of the annotation should be documented and considered when interpreting search results.

### Variant Representation

The representation of variants in the database is limited by the variant detection method. Variants that are not detected in the DNA or RNA sequencing data cannot be included in the database.

Low-frequency variants may be missed by variant calling, and structural variants may not be represented by simple amino acid substitutions. The database should be constructed with the understanding that it represents the detectable variant landscape, not the complete variant landscape.

### Immunoglobulin Diversity

Immunoglobulin and T-cell receptor sequences present a special challenge because of their extreme diversity. Standard database construction approaches are insufficient for these sequences.

The antibody repertoire study demonstrated that specialized approaches using graph structures can capture the diversity of immunoglobulin genes in a compact manner. These approaches require specialized software and expertise.

## Safety and Regulatory Context

### Data Usage Agreements

Genomic and transcriptomic data downloaded from public repositories may be subject to data usage agreements. Some datasets, particularly those derived from human subjects, may have restrictions on their use.

Researchers should review the data usage agreements for all data sources before constructing a database. The agreements should be documented in the provenance record.

### Ethical Considerations

The use of human genomic data for database construction raises ethical considerations. The data should be used in accordance with the consent agreements under which they were collected.

Researchers should ensure that their use of human data is approved by the appropriate institutional review board or ethics committee.

### Reproducibility Requirements

Many journals now require that proteomics data and analysis pipelines be deposited in public repositories. The database construction pipeline should be documented in sufficient detail to allow others to reproduce the analysis.

The [nf-core](https://nf-co.re/docs) documentation provides guidance on creating reproducible analysis pipelines that meet community standards.

## Professional Escalation Criteria

### When to Seek Specialized Assistance

Database construction can become complex, and there are situations where specialized assistance is warranted. These situations include working with organisms that have no close relatives in public databases, analyzing immunoglobulin repertoires, constructing databases for metaproteomics studies with many organisms, and incorporating complex structural variants.

The review of metaproteomics database construction provides practical recommendations for constructing databases and discusses the current challenges in this area. Researchers facing these challenges should consult the review literature and consider collaborating with bioinformatics specialists.

### When to Consult the Literature

The scientific literature provides guidance for specific database construction scenarios. The antibody repertoire study provides a detailed protocol for constructing immunoglobulin databases. The metaproteomics review provides recommendations for constructing databases for microbial communities.

Researchers should search the literature for protocols relevant to their specific organism or application before developing a custom approach.

### When to Escalate to a Core Facility

Many institutions have bioinformatics core facilities that provide database construction services. These facilities have expertise in the tools and workflows described here and can provide guidance on best practices.

Researchers who lack the computational expertise or resources to construct a database should consider engaging their institutional core facility. The core facility can also provide guidance on data management and reproducibility.

## Frequently Asked Questions

### What is the minimum information needed to build a custom protein sequence database?

The minimum information is the genome assembly or transcriptome for the organism of interest, plus the annotation file if available. For genome-based construction, the genome FASTA file and the GFF or GTF annotation file are required. For transcriptome-based construction, the assembled transcript sequences are required. The genetic code for the organism must also be known.

### How do I choose between genome-based and transcriptome-based database construction?

Genome-based construction is appropriate when a high-quality genome assembly and annotation are available. Transcriptome-based construction is appropriate when RNA-seq data from the sample or a closely related sample are available, because it captures sample-specific expression and splice isoforms. For non-model organisms without annotation, transcriptome-based construction is often more practical than genome-based prediction.

### What is the difference between target and decoy sequences?

Target sequences are the protein sequences that are expected to be present in the sample. Decoy sequences are reversed or shuffled versions of the target sequences that are used to estimate the false discovery rate. The search database contains both target and decoy sequences, and matches to decoy sequences are assumed to be false positives.

### How many contaminant sequences should I include?

The contaminant database should include all proteins that are expected to be present in the sample from handling and reagents. A standard contaminant database typically contains 100 to 300 sequences, including keratins, serum proteins, and digestion enzymes. The exact number depends on the laboratory environment and sample preparation methods.

### Can I add variants to an existing reference proteome?

Yes, variants can be added to an existing reference proteome by modifying the reference sequences and adding the modified sequences as separate entries. Each variant sequence should have a header that indicates the reference sequence and the variant change. This approach preserves the reference sequences while adding the variant information.

### How do I know if my database is too large?

A database is too large when the search time becomes impractical or when the false discovery rate becomes unacceptably high. The acceptable database size depends on the computational resources and the search engine used. As a general principle, the database should contain all relevant sequences and as little irrelevant sequence as possible.

### What should I do if my search produces no identifications?

If a search produces no identifications, the first step is to verify that the database is in the correct format and contains the expected sequences. The next step is to run a test search against a well-characterized sample to verify that the search pipeline is working. If the test search works, the problem may be in the database content or in the sample preparation.

### How should I document my database construction for publication?

The documentation should include the source data accessions and versions, the tool versions and parameters, the processing steps, and the quality control results. This information should be included in the methods section of the publication and in a supplementary file. The [nf-core](https://nf-co.re/docs) documentation provides guidance on creating reproducible analysis pipelines that meet community standards.

## Related Bioinformatics Guides

- [RNA-Seq Databases: Accessing and Using Public RNA-Seq Data](/knowledge/bioinformatics/rna-seq-databases-accessing-and-using-public-rna-seq-data)
- [Volcano Plot Proteomics: How to Create and Interpret Them Effectively](/knowledge/bioinformatics/volcano-plot-proteomics-how-to-create-and-interpret-them-effectively)
- [Mass Spectrometry Protein Identification: From Raw Spectra to Confident Hits](/knowledge/bioinformatics/mass-spectrometry-protein-identification-from-raw-spectra-to-confident-hits)
- [Proteomics Data Analysis Workflow: From Raw Spectra to Biological Insights](/knowledge/bioinformatics/proteomics-data-analysis-workflow-from-raw-spectra-to-biological-insights)
- [Genomic Data Processing: From Raw Sequencing to Analysis-Ready Files](/knowledge/bioinformatics/genomic-data-processing-from-raw-sequencing-to-analysis-ready-files)

## 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.
- [Construction of Protein Sequence Databases for Metaproteomics: A Review of the Current Tools and Databases.](https://pubmed.ncbi.nlm.nih.gov/39449618). Journal of proteome research, 2024.
- [FP-Zernike: An Open-source Structural Database Construction Toolkit for Fast Structure Retrieval.](https://pubmed.ncbi.nlm.nih.gov/38894604). Genomics, proteomics & bioinformatics, 2024.
- [The Antibody Repertoire of Colorectal Cancer.](https://pubmed.ncbi.nlm.nih.gov/29046389). Molecular & cellular proteomics : MCP, 2017.
- [Proteomics-based prognostic signature in colon adenocarcinoma patients with familial adenomatous polyposis.](https://pubmed.ncbi.nlm.nih.gov/40687235). Translational cancer research, 2025.
- [Construction of a two-dimensional gel electrophoresis protein database for the Nicotiana tabacum cv. Bright Yellow-2 cell suspension culture.](https://pubmed.ncbi.nlm.nih.gov/14997494). Proteomics, 2004.

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