# Gene vs. Nucleotide vs. Protein A gene is the curated biological unit, a nucleotide record is a specific DNA or RNA sequence, and a protein record is the amino acid chain that a coding sequence encodes. In NCBI terms, the Gene database tells you what is known about a locus, the Nucleotide database hands you the sequence molecules, and the [Protein database](/blog/guides/protein-database) hands you the translated products and their features. The rule of thumb is simple. Start at NCBI Gene when you begin from a gene name and need context. Start at Nucleotide when you begin from an accession number or need DNA letters. Start at Protein when you begin from amino acids or care about domains, structure, and what a variant does to the encoded product. These three levels describe the same information at different resolutions, and most confusion in public sequence databases comes from working at the wrong level. A primer designed against the wrong transcript, a variant called against the wrong reference, or a BLAST run against the wrong database all trace back to the same root cause. This guide is a practical NCBI databases guide for the ncbi gene vs nucleotide vs protein question. It explains what each database holds, how records link to each other, which entry point fits which task, and how published studies move between the three levels in real workflows. First, the vocabulary. A gene is a stretch of DNA that is transcribed into RNA and, for protein-coding genes, translated into protein, together with the regulatory signals that control it. A nucleotide is one unit of DNA or RNA, the bases adenine, cytosine, guanine, and thymine (uracil replaces thymine in RNA), and database shorthand uses "nucleotide record" to mean a stored sequence of those units. A protein is a chain of amino acids that folds into a working molecule. The central dogma of [molecular biology](/blog/careers/molecular-biology), DNA to RNA to protein, is also the data model behind NCBI's three databases. ## NCBI gene vs nucleotide vs protein at a glance | Question | NCBI Gene | NCBI Nucleotide | [NCBI Protein](/blog/guides/ncbi-protein) | |--|--|--|--| | What one record is | A locus, the gene as a curated object | One DNA or RNA sequence with annotations | One [amino acid sequence](/blog/guides/amino-acid-sequence) with annotations | | Identifier | GeneID, a stable integer | Versioned accession, such as MG385063.1 or NM_144701.3 | Versioned accession, such as an NP_ or XP_ number, or a UniProt ID like Q5VWK5 | | Best starting point | You have a gene name or symbol | You have an accession number or raw DNA | You have amino acids or a protein name | | What you get | Function, phenotype, map position, orthologs, linked sequences and variants | The sequence itself, CDS coordinates, UTRs, RNA features | The sequence, domains, sites, structures, and paired transcripts | | Letters stored | None, the record links out | A, C, G, T (or U for RNA) | The 20 amino acid codes | | BLAST program | Not applicable | blastn | blastp | | Change behavior | GeneID stays stable while content is updated | Version increments when the sequence changes | Version increments, paired to transcript versions | | Typical use | Decide which transcript and records to work with | Primer design, cloning, submission, alignment baselines | Domain and structure analysis, variant interpretation | Accession formats follow conventions worth memorizing. Traditional GenBank nucleotide accessions are one letter plus five digits or two letters plus six digits, for example MG385063, KY706080, and CP069561, three accessions used as reference baselines for antibiotic resistance genes in an Escherichia coli study [1]. Protein accessions are typically three letters plus five digits. RefSeq, the curated reference collection, uses prefixes: NM_ for curated messenger RNA, NR_ for noncoding RNA, XM_ for predicted transcript, NP_ for curated protein, XP_ for predicted protein, NC_ for a complete genomic molecule, and NG_ for a reference genomic locus. Every accession carries a version suffix written as accession.version, such as NM_144701.3. The version number increments whenever the sequence letters change, so citing the version pins the exact sequence you analyzed. ## What each NCBI database actually contains ### The Gene database: a curated entry point, not a sequence archive The Gene database describes loci, not sequences. Each record centers on a GeneID, a stable integer that survives renaming, reannotation, and transcript revisions. Around that ID, the record collects the preferred symbol and its aliases, the organism and genomic map position, a summary of function, links to RefSeq transcripts and proteins, phenotype descriptions, pathway assignments, interaction partners, and orthologs, which are the corresponding genes in other species. Gene records also cover predicted genes and pseudogenes, so the record status is worth checking. Gene nomenclature is messy in the wild. The same symbol can refer to different loci in different organisms, symbols get renamed, and one gene can carry several aliases. The GeneID exists to cut through that ambiguity, and that is the core reason to use NCBI Gene as the anchor for a project. In practice, you reach for Gene when you want to do the following. 1. Anchor a project to a stable identifier that does not change when symbols do. 2. Find the reference transcript set before designing primers or calling variants. 3. Gather phenotype, pathway, and interaction context that no single sequence record contains. 4. Locate orthologs before cross-species work. 5. Reach variation data. Gene records connect to dbSNP, NCBI's archive of short genetic variation, where each variant carries an rs number. The context layers matter. The RET study drew 28,335 SNPs from NCBI dbSNP to build its variant catalog [3]. The CD40 meta-analysis used NCBI resources to profile mutant allele distributions for four SNPs across six global populations, then layered tumor expression and survival data from GEPIA and an interaction network from STRING onto the same gene [4]. None of that context lives in a sequence record. It lives in the gene-centered layer that Gene organizes. ### The Nucleotide database: GenBank, RefSeq, and versioned accessions The Nucleotide database stores DNA and RNA sequences. Two families live side by side. GenBank is the archival collection of author-submitted sequences, exchanged daily with its partners in the International [Nucleotide Sequence](/knowledge/molecular-biology/nucleotide-sequence) Database Collaboration, the European Nucleotide Archive and the DNA Data Bank of Japan. RefSeq is NCBI's curated, non-redundant reference set, built from those submissions and from genome assemblies and maintained by NCBI staff. Every nucleotide record carries a feature table that turns raw letters into biology. The source feature names the organism. Gene features mark loci. The CDS feature, short for coding sequence, marks the exact interval that translates into protein and carries a /translation qualifier containing the amino acid string. Exon features mark spliced segments. The regions flanking the CDS are the UTRs, the untranslated regions that shape translation efficiency and, frequently, microRNA binding. Not everything in Nucleotide encodes a protein. Introns, regulatory regions, and the large catalog of noncoding RNAs live here too, and some of the most informative analyses never touch a protein record. One study of human microRNA-mRNA interactions mined 2,568 miRNAs from the NCBI database against the mRNAs of 17,508 protein-coding genes and found binding sites in the mRNAs of 384 genes for 89 miRNAs, with most sites located in 3' UTRs [5]. That regulatory layer exists only at the nucleotide level, which is a standing argument for reading records beyond their coding intervals. ### The Protein database: amino acid records with functional features The Protein database stores amino acid sequences. It aggregates RefSeq proteins (NP_ and XP_ records), curated Swiss-Prot entries from UniProtKB, and protein chains from the PDB structure archive. A protein record gives you more than letters. Feature annotations mark domains, active sites, disulfide bonds, and transmembrane segments, and links lead to three-dimensional structures and to the transcripts that encode the protein. Protein-level questions are different questions. What does this sequence do? Which domains does it carry? Does a variant land inside a catalytic or binding domain? A sugarcane study shows the pattern. Degenerate primers amplified candidate resistance sequences, 1,512 positive clones assembled into 134 contigs (a contig is a contiguous assembled sequence), and the contig consensuses were compared against the NCBI sequence database using BLASTx. Twenty contigs showed homology to NBS-LRR resistance gene analogues and eight to kinase genes, and alignment of the deduced amino acid sequences pinned down the conserved domains [6]. The decisive evidence was protein-level, because domains, not individual nucleotides, carry the functional signal. ## How gene, nucleotide, and protein records connect ### From a Gene record to its sequences A Gene record is a hub. Its transcript table lists every RefSeq transcript paired with the protein it encodes, flags the reference transcript, and separates curated models from predicted ones. Choosing the transcript is a real decision, because isoforms, the alternative splice products of one gene, can differ in exons, UTR length, and coding content. Primer design, variant calling, and expression assays all inherit that choice, so make it deliberately rather than accepting the first record on the page. ### Coding sequences and the reading frame The reading frame is the phase in which nucleotides are grouped into codons, the triplets that specify amino acids. Shifting the frame by one or two bases changes every downstream codon, so the same nucleotide segment read in a different frame yields a completely different protein. That is why CDS features are frame-locked, with coordinates and a codon_start qualifier that fix the phase. Annotation is not always consistent. Multiple annotation sources often coexist for the same genome, from prediction pipelines to manual curation to independent annotations from public repositories such as NCBI, and frame-blind comparisons can miss conflicts that change the protein. The CDScompare tool was built for exactly this problem, performing frame-sensitive comparison of GFF files (the General Feature Format, a tabular annotation standard) and flagging coding discrepancies in wheat and rice annotations [7]. When you extract a CDS from a record, keep the annotated coordinates and phase, and do not re-derive the frame by eye. ### Transcript-protein reference pairs RefSeq transcripts and proteins are versioned in pairs, so a transcript accession such as NM_144701.3 corresponds to one specific protein product. Anchoring every analysis to one named pair is what makes variant interpretation reproducible. The IL23R study is a clean example. It took 723 missense SNVs from NCBI dbSNP, verified each one against transcript NM_144701.3 and its 629 amino acid protein (UniProt Q5VWK5-1) using Ensembl VEP on the GRCh38 assembly, then filtered them through SIFT, PolyPhen-2, PROVEAN, E-SNPs&GO, MutPred2, and ConSurf, with AlphaMissense and FATHMM-MKL as independent annotation layers. Thirty-seven high-risk variants survived, and 36 of 37 were predicted to destabilize the protein [2]. Every number in that pipeline depends on the named transcript version. Change the transcript and the variant list changes with it. ### Variant names that span all three levels Variant nomenclature encodes the level by prefix. A c. name gives position in coding DNA, such as c.902C>A. A p. name gives the protein consequence, such as p.Ser301STOP. The DMRT3 gene in horses, known as the gait keeper, shows all three levels in one locus. The stop-gain variant c.902C>A (p.Ser301STOP) turns codon 301 from serine into a stop signal and truncates the protein, and it is the variant tied to gaited movement in Icelandic Horses and French Trotters. A novel 18 nucleotide in-frame duplication, c923_943dup (Ala297_Ala302dup), found in Shetland Ponies, inserts six amino acids without disturbing the frame. Both were annotated against Ensembl and NCBI records [8]. Two adjectives classify coding substitutions. A synonymous change swaps one codon for another codon that specifies the same amino acid, so the protein is unchanged. A missense (non-synonymous) change specifies a different amino acid. In exon 1 of the buffalo MC1R gene, an 822 bp region (bp, base pair, the unit of nucleotide length) resequenced across color variants, five variation sites were identified, one synonymous and the rest non-synonymous and predicted to alter the amino acid sequence of the MC1R protein [9]. Whether a nucleotide substitution is silent or consequential is decided at the codon level, which is exactly the boundary between the Nucleotide and Protein views. ## How to use NCBI without getting lost Entrez, NCBI's unified search engine, queries all databases at once and shows result counts per database, which is the fastest way to see whether your term is primarily a gene name, a sequence, or a publication. Three habits prevent most problems after that. First, use the links rather than new searches. A Gene record links to its transcripts, proteins, variation, pathways, and literature. Following links keeps every item tied to the same locus and prevents the quiet mix-up of pulling a transcript from the wrong species or the wrong isoform. Second, control your downloads. The Send to menu on any search or record exports selected records as FASTA, the plain text format that stores a sequence after a one-line header beginning with >. For scripted work, the E-utilities fetch records directly. A GenBank flat file for a nucleotide accession comes from efetch: `https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=nucleotide&id=MG385063&rettype=gb&retmode=text` The same endpoint with db=protein and rettype=fasta returns a protein in FASTA format. The rettype=gb output is the annotated flat file with the feature table, CDS coordinates, and /translation qualifier described above. Third, read the record before you trust it. Check the organism, the molecule type, the version, and whether the record is curated RefSeq or a submitted GenBank entry. Those four fields decide whether the record can serve as a reference baseline for your analysis. ## When to use NCBI Gene, Nucleotide, or Protein ![Decision tree that starts from what you have in hand and routes to Gene, Nucleotide, or Protein, then to the matching BLAST program.](/article-images/illustrations/ncbi-gene-vs-nucleotide-vs-protein-entry-point-choice.webp) ### Start at Gene when you have a name or need context You typed a symbol from a paper or a protocol. Search Gene first, pick the organism, and read the record. You will leave with the right transcript accessions, the orthologs you will need later, and the phenotype context that tells you whether the gene does what you assume. Gene is also the correct entry point when the deliverable is understanding rather than a file of letters. ### Start at Nucleotide when you have an accession or need DNA An accession number from a paper, a sequence you need to order or clone, a reference to align against, or a batch of sequences to download for population analysis all start in Nucleotide. This is also the database you submit new sequences to through NCBI's submission routes, which makes it the endpoint of most sequencing projects as well as the starting point of the next analysis. ### Start at Protein when you need amino acids, domains, or structure An amino acid query, a domain search, a structure question, or a variant that needs interpretation in terms of protein features all start in Protein. If your downstream tools expect amino acids, work at this level and return to Nucleotide only when you need the codons. ### Picking the right BLAST program BLAST, the Basic Local Alignment Search Tool, finds locally similar sequences in a database. The program name encodes the query type and the target type. | Your query | Search against | Program | Typical question | |--|--|--|--| | Nucleotide | Nucleotide | blastn | Which known sequences match my DNA? | | Protein | Protein | blastp | Which proteins are homologous to mine? | | Nucleotide | Protein | blastx | What proteins could my DNA encode in any frame? | | Protein | Nucleotide | tblastn | Which DNA records encode a protein like mine? | | Nucleotide | Nucleotide (both translated) | tblastx | Is there distant coding similarity between two DNAs? | The sugarcane study used blastx because the query was nucleotide (contig consensus sequences) but the meaningful comparison was at the protein level, where conserved domains remain detectable after nucleotide similarity has decayed [6]. The general principle holds across projects. Protein comparisons detect more distant homology than nucleotide comparisons because the genetic code is degenerate, meaning several codons specify the same amino acid. ## Worked workflows ### Workflow 1: From a gene symbol to its variant catalog Task: you have a gene symbol and want to know which variants matter. 1. Search Gene for the symbol and pick the organism. 2. Open the variation links to dbSNP and filter to missense variants on the reference transcript. 3. Map surviving variants onto protein domains. 4. Score them with prediction tools and, where structures exist, model them. The RET study followed exactly this path. It retrieved 28,335 SNPs from NCBI dbSNP, including 2,377 nonsynonymous SNPs, screened them with ten in silico prediction tools, and converged on 33 high-risk variants, 28 of them predicted to destabilize the protein. Eleven potentially pathogenic nsSNPs (R721G, A756D, Y791C, E734K, E805K, F893L, R897Q, R897G, R912Q, R912G, and M918T) clustered in the tyrosine kinase domain, and [molecular docking](/knowledge/bioinformatics/docking-algorithms-autodock-glide-and-beyond) (computational prediction of how a molecule binds a protein) showed entrectinib holding strong binding affinities across variants, from -9.7 to -10.7 kcal/mol [3]. The ING1 study shows the same pipeline shape with heavier structural machinery. It started from 347 missense nsSNPs in dbSNP, screened them with twelve sequence-based tools, built a three-dimensional model with AlphaFold3, refined it with GalaxyRefine, validated it with ERRAT, PROCHECK, and TM-align, and scored mutation effects with Missense3D, mCSM, and BeAtMuSiC before mapping interactions in STRING and GeneMANIA [10]. Association studies add a population layer. A Korean prostate cancer study selected 21 SNPs in CYP24A1 from the HapMap and NCBI databases using minor allele frequency and [linkage disequilibrium](/knowledge/bioinformatics/linkage-disequilibrium-and-haplotype-mapping) information (the nonrandom co-inheritance of alleles at nearby sites), genotyped 272 cases and 173 controls, and reported five variants (rs2248461, rs2248359, rs6022999, rs2585428, and rs4809959) associated with risk after false discovery rate correction, with odds ratios between 0.46 and 0.65 [11]. The database step, choosing which SNPs to genotype, is what makes the laboratory step efficient. ### Workflow 2: From an accession number to a reference baseline ![Flowchart of the accession workflow: fetch the record, check version and organism, locate the CDS, translate, then classify each substitution.](/article-images/illustrations/ncbi-gene-vs-nucleotide-vs-protein-accession-to-baseline.webp) Task: you sequenced a gene and need to compare it with a published reference. 1. Enter the accession in the Nucleotide search box. 2. Check the version and the organism. 3. Read the feature table and locate the CDS. 4. Extract the CDS, translate it, and align your sequence. 5. Call substitutions and classify each as synonymous or missense. An Escherichia coli study of urinary tract infection isolates worked exactly this way against GenBank references. The aadA1 sequences showed 10 nucleotide substitutions compared with reference MG385063, 9 of them with missense effects. The dfrA1 sequences showed 20 substitutions against KY706080, 8 missense. The sul1 sequences showed 20 substitutions against CP069561, 12 missense [1]. The reference accessions carry the meaning. Without a versioned baseline, a count of substitutions has no anchor, and with one, every downstream claim is checkable by anyone who fetches the same record. ### Workflow 3: Designing and validating PCR primers Task: you need primers that amplify one target and nothing else. 1. Retrieve the target locus or transcript from Nucleotide. 2. Design primers and check them against the whole genome with NCBI Primer-BLAST, which tests candidate primer pairs for unintended binding. 3. Validate in silico, then in the laboratory. A Salmonella Typhi study designed primers against the invA, phsB, and tviA genes, optimized thermal profiles by conventional PCR, then moved to SYBR Green real-time PCR (a fluorescence-based quantitative PCR chemistry) for rapid detection. The invA marker amplified in 90% of Salmonella isolates, supporting its use as a genus-level marker, and Sanger sequencing of the invA amplicon confirmed that the region is conserved [12]. A Vibrio vulnificus study shows the failure mode and the fix. Published primers for vvhA and rpoS gave inconsistent results across environmental isolates, so the authors retrieved whole genome sequences through NCBI Primer-BLAST for the reference strain ATCC 27562 and designed new primer sets with in silico specificity checks [13]. The nucleotide database supplies the raw material and Primer-BLAST narrows the risk, but laboratory validation remains the gate. ### Workflow 4: Comparing sequences across breeds, species, and populations Task: you want to know how a locus varies across groups. 1. Retrieve homologous sequences from Nucleotide, one per sample or species. 2. Align them and compute diversity statistics. 3. Interpret at both the nucleotide and protein levels. A phylogeography study of avian Isospora retrieved 273 mitochondrial COI sequences from NCBI spanning 17 countries. The alignment yielded 119 haplotypes (a haplotype is a specific combination of sequence variants inherited together) with high haplotype diversity (0.944) and low nucleotide diversity (0.009). Tajima's D (-2.646) and Fu's Fs (-203.014), two neutrality tests whose negative values signal an excess of rare variants, were consistent with recent population expansion, and analysis of molecular variance attributed 95.26% of the variation to differences within populations [14]. The livestock example teaches the interpretive lesson. The HSP70.1 heat shock gene has a 1,926 bp coding region in Indian zebu cattle, Bos taurus, and riverine buffalo alike, encoding a 641 amino acid protein of 70.26 kDa (kilodaltons, a unit of molecular mass). The UTRs are a different story. Buffalo carried longer 5' and 3' UTRs (204 and 293 nucleotides) than taurine and indicine cattle (172 and 286 nucleotides), owing to insertions, and 54 variations (50 SNPs and 4 indels) separated the species [15]. A protein-level comparison would call these genes equivalent. The nucleotide-level comparison reveals regulatory architecture that differs, and UTR length can matter for transcript stability and microRNA access. A native chicken study closes the loop from reference to association. Primers were designed for the GNRHR gene, amplicons from 100 Isfahani native hens were sequenced, the sequences were blasted against the reference gene in NCBI, and five SNPs were called at nucleotide positions 117, 123, 157, 453, and 550 of the amplified fragment (T to C, C to T, C to T, C to T, and A to G) before genotypes were tested against production traits [16]. Reference record, resequencing, alignment, variant calling, association. That loop is the everyday workhorse of comparative genetics, and every step is anchored to a database record. ### Workflow 5: From a nucleotide edit to a protein consequence Task: you want to edit a gene and predict the protein outcome before touching a cell. 1. Retrieve the gene from NCBI and choose the target exon. 2. Pick the codon to edit and compute the codon-level consequence. 3. Design donor DNA and validate the construct by sequencing. A [genome editing](/blog/guides/genome-editing) study in rohu (Labeo rohita) targeted mstnb, a negative regulator of skeletal muscle growth. The gene sequence was retrieved from the NCBI database, a target site was chosen in exon 1, and a single G to T substitution was planned to convert the glycine codon GGA into the stop codon TGA, introducing a premature stop upstream of the native stop signal. A 600 bp donor fragment carrying the edit, flanked by recombinase target sites, was amplified, cloned into the pJET1.2 vector, and confirmed by Sanger sequencing, with protein consequences assessed using InterPro and ColabFold [17]. Every decision crossed levels. The gene defined the target, the nucleotide sequence defined the edit, and the protein prediction defined the expected effect. Codon arithmetic, one base change turning a glycine codon into a stop signal, is the clearest demonstration that the nucleotide and protein views are two readings of the same molecule. ## Common Mistakes and Limitations **Treating a GeneID and an accession as interchangeable.** A GeneID names a locus. An accession names one sequence record, often one transcript or one submitted fragment. A gene with twelve transcripts has twelve mRNA accessions and one GeneID. Cite the one you mean. **Citing an accession without its version.** NM_144701.3 and a future NM_144701.4 are not the same letters. The version suffix is part of the citation, and analyses that omit it are hard to reproduce. **Using predicted records as if they were curated.** XM_ and XP_ records are computational predictions. They are useful, and they are not the same commitment as NM_ and NP_ records. Check the prefix before you build a primer or call a variant against a predicted model. **Ignoring the frame and annotation conflicts.** The same nucleotide segment in a different frame encodes a different protein, and multiple annotation sources routinely coexist for the same genome [7]. When two sources disagree about a CDS, one of them is wrong, and frame-aware comparison is how you find out. **Assuming the database is complete.** Resequencing of the CYP1A1 and CYP1A2 region in 24 subjects found 85 SNPs, 49 of which were absent from the NCBI database at the time, and the variants most used in epidemiology turned out not to be representative tag SNPs (variants that stand in for a block of correlated variants) across three human subgroups [18]. Databases lag the world's variation. Absence from a database is not evidence of absence in a population. **Reading annotation as experimental fact.** Feature tables mix experiment, computation, and transfer from related organisms. The studies above that carried real-world weight all closed the loop at the bench, with Sanger sequencing confirming conserved invA amplicons [12] and confirming the edited mstnb construct [17]. Every variant-effect score in the workflows is a computational prediction, a hypothesis that needs confirmation for any individual case, and sequence findings that will inform clinical or experimental decisions deserve expert interpretation and laboratory confirmation. **Running the wrong BLAST program.** A nucleotide query against the protein database does nothing until it is translated, and a protein query against the nucleotide database is wasted unless the program translates the target. Match the program to the letters you have, as in the BLAST table above. **Forgetting that one gene has many transcripts.** Primer design against the wrong isoform is a classic silent failure. Decide the transcript first, at the Gene record, then design. **Assuming identical proteins mean identical genes.** The HSP70.1 comparison showed species with the same coding region length and protein size but different UTR architecture [15]. Protein identity is not nucleotide identity, and regulation lives in the parts of the gene that never appear in a protein record. ## Frequently Asked Questions ### What is the difference between NCBI Gene, Nucleotide, and Protein? Gene describes loci, Nucleotide stores DNA and RNA sequences, and Protein stores amino acid sequences. A Gene record tells you what is known about a gene and links to its sequences, a Nucleotide record gives you the sequence itself with its annotations, and a Protein record gives you the encoded chain with its domains and features. ### When should I use NCBI Gene instead of searching Nucleotide first? Use Gene when you start from a gene name or symbol, or when you need context such as the reference transcript set, phenotypes, pathways, and orthologs. Use Nucleotide when you already have an accession number or a sequence in hand. ### Is a GenBank accession number the same thing as a gene? No. An accession identifies one sequence record, which may be a fragment, a single transcript, or a complete genome. A gene is a locus that can span many records, and the stable handle for the locus is the GeneID. ### How do I get the protein sequence from a nucleotide record? Open the record's CDS feature and use its /translation qualifier, or click the linked protein accession in the feature table. RefSeq transcript records pair with RefSeq protein records, so an NM_ transcript points to its NP_ protein. ### Which BLAST program should I use? blastn for a nucleotide query against a nucleotide database, blastp for protein against protein, blastx for a nucleotide query translated and searched against proteins, and tblastn for a protein query against a translated nucleotide database. Match the program to the letters you have. ### Are RefSeq records better than GenBank records? They serve different purposes. RefSeq records are curated, non-redundant references stable enough to anchor analyses. GenBank records are the archival author-submitted record from which much of RefSeq is built. Reference work usually starts with RefSeq, while submission, novelty checking, and archival retrieval use GenBank. ## Sources 1. [Analysis of Nucleotide Sequences Similarity and Protein Prediction of Some Resistance Genes in Escherichia coli Isolated from Iraqi Patients with Urinary Tract Infections.](https://pubmed.ncbi.nlm.nih.gov/36618275/) 2. [In Silico Identification and Structural Characterization of High-Risk Missense SNVs in the Human IL23R Gene Relevant to Inflammatory Bowel Disease.](https://pubmed.ncbi.nlm.nih.gov/42353858/) 3. [Modeling the Effects of Single Nucleotide Polymorphisms (SNPs) on the Structure and Function of the Human RET Gene: An In Silico Study.](https://pubmed.ncbi.nlm.nih.gov/42369210/) 4. [The association between CD40 gene polymorphisms and the risk of cancer based on a meta-analysis.](https://pubmed.ncbi.nlm.nih.gov/42245400/) 5. [Fully Complementary Interactions of Human mRNA Genes with miRNAs from NCBI Database.](https://pubmed.ncbi.nlm.nih.gov/42939514/) 6. [Isolation of nucleotide binding site-leucine rich repeat and kinase resistance gene analogues from sugarcane (Saccharum spp.).](https://pubmed.ncbi.nlm.nih.gov/17935262/) 7. [CDScompare: a frame-sensitive GFF comparison tool for improved detection of coding sequence discrepancies, with applications to wheat and rice annotations.](https://pubmed.ncbi.nlm.nih.gov/42958397/) 8. [DMRT3 gene variations in horse breeds selected for gaited movement - established research and novel findings.](https://pubmed.ncbi.nlm.nih.gov/41936970/) 9. [Novel mutation sites in the Bubaline MC1R gene.](https://pubmed.ncbi.nlm.nih.gov/41923858/) 10. [Predicting the Impact of Deleterious Single-Nucleotide Polymorphisms in the p47ING1a Isoform of Human ING1 Gene.](https://pubmed.ncbi.nlm.nih.gov/42322085/) 11. [Genetic variants in the CYP24A1 gene are associated with prostate cancer risk and aggressiveness in a Korean study population.](https://pubmed.ncbi.nlm.nih.gov/24492489/) 12. [Evaluation of gene markers invA, phsB, and tviA for detection of Salmonella Typhi by SYBR Green (TM) real-time PCR assay in clinical samples.](https://pubmed.ncbi.nlm.nih.gov/42584516/) 13. [Comparison of primers for the pathogenicity factors vvhA and rpoS in Vibrio vulnificus environmental isolates from the Texas Coastal Bend region of the Gulf of Mexico.](https://pubmed.ncbi.nlm.nih.gov/41943823/) 14. [Phylogeography and genetic structure of avian isospora based on mitochondrial COI gene sequences from 17 countries.](https://pubmed.ncbi.nlm.nih.gov/42102985/) 15. [Novel polymorphisms in UTR and coding region of inducible heat shock protein 70.1 gene in tropically adapted Indian zebu cattle (Bos indicus) and riverine buffalo (Bubalus bubalis).](https://pubmed.ncbi.nlm.nih.gov/23792016/) 16. [Polymorphism of the Gonadotropin-Releasing Hormone Receptor Gene and Its Association With Production Traits in Isfahani Native Hens.](https://pubmed.ncbi.nlm.nih.gov/42206885/) 17. [Design and in silico validation of donor DNA for RNA-guided recombinase-mediated knockout of mstnb gene in Labeo rohita.](https://pubmed.ncbi.nlm.nih.gov/42455869/) 18. [Toward the evaluation of function in genetic variability: characterizing human SNP frequencies and establishing BAC-transgenic mice carrying the human CYP1A1_CYP1A2 locus.](https://pubmed.ncbi.nlm.nih.gov/15643613/)