Bioinformatic Resources: Sequence, Expression, Structure and Pattern Databases
Primary and secondary biological databases for CSIR NET: INSDC, RefSeq, UniProt, GEO, PDB, PROSITE, Pfam, InterPro, GO, file formats and Phred scores.
- 19 min
On this page · 16 sections
- Prerequisites
- The Hook: One Gene, Twenty Records
- What a Biological Database Is
- Sequence Databases
- Expression Databases
- Structure Databases
- Pattern, Motif and Family Databases
- Ontologies, Pathways and Interactions
- File Formats You Must Recognise
- Memory Tricks
- Don’t Confuse
- Common Mistakes
- Practice
- Quick Revision Table
- Key Takeaways
- What’s Next
Prerequisites
- Carbohydrates, Lipids, Nucleic Acids and Vitamins — what a nucleotide sequence is before you store one.
- Protein Structure, Conformation and Stability — motifs, domains and folds, which the pattern and structure databases catalogue.
- Transcription and RNA Processing — mRNA, splicing and isoforms explain why one gene has many database records.
The Hook: One Gene, Twenty Records
How CSIR NET asks it. This is a new unit in the revised LS-BT syllabus, so there are no past CSIR papers to go by. Expect Part B to ask matching questions (database to data type, accession prefix to molecule type, Swiss-Prot vs TrEMBL), and Part C to ask you to pick the right resource for a stated research task, decode a FASTQ quality string, or work out sequencing coverage.
What a Biological Database Is
A biological database is an organised, searchable collection of biological data with a defined record structure, unique identifiers (accession numbers), cross-references to other databases, and a query interface. The first organised collection is usually traced to Margaret Dayhoff’s Atlas of Protein Sequence and Structure in the 1960s, which later gave us the PAM matrices (Sequence Alignment).
Primary, secondary and specialised
| Type | What it holds | Who writes the record | Examples |
|---|---|---|---|
| Primary (archival) | Raw experimental data: sequences, structures, expression data, as submitted | The submitting laboratory | GenBank, ENA, DDBJ, SRA, PDB, GEO, ArrayExpress |
| Secondary (derived / curated) | Information computed or curated from primary data: patterns, families, annotation, non-redundant sets | Curators and algorithms | UniProtKB/Swiss-Prot, RefSeq, PROSITE, Pfam, PRINTS, InterPro, SCOP, CATH |
| Specialised / composite | One organism, one disease, one data type, or several sources merged | Mixed | FlyBase, SGD, TAIR, OMIM, KEGG, STRING, PubChem |
Primary archives are redundant (the same gene can be submitted many times) and the submitter “owns” the record: the database does not correct the biology. Curated secondary resources remove redundancy, correct and add annotation, and attach evidence.
graph LR
EXP["Experiment
sequencing, crystallography,
microarray, RNA-seq"] --> P["Primary archives
GenBank/ENA/DDBJ, SRA,
PDB, GEO"]
P --> S["Secondary resources
RefSeq, UniProtKB,
Pfam, PROSITE, CATH"]
S --> I["Integrated portals
Entrez, Ensembl,
InterPro, UniProt"]
P --> I
I --> U["Researcher
query, download, analyse"]Sequence Databases
Nucleotide: the INSDC
The International Nucleotide Sequence Database Collaboration (INSDC) has three partners that exchange data every day, so a sequence submitted to one appears in all three under the same accession number:
| Partner | Host | Country |
|---|---|---|
| GenBank | NCBI (National Center for Biotechnology Information), part of the US National Library of Medicine | USA |
| ENA (European Nucleotide Archive, formerly the EMBL Nucleotide Sequence Data Library) | EMBL-EBI, Hinxton | UK |
| DDBJ (DNA Data Bank of Japan) | National Institute of Genetics, Mishima | Japan |
INSDC also covers raw reads: NCBI’s SRA (Sequence Read Archive), the ENA read archive and the DDBJ read archive. Accessions are versioned: in AB123456.2 the .2 means the sequence has been revised once. A changed sequence gets a new version; the accession stays.
RefSeq: the curated reference set
NCBI’s RefSeq is a non-redundant, curated set: ideally one record per natural molecule per organism. Its accession prefixes are high-yield.
| Prefix | Molecule | Status |
|---|---|---|
| NC_ | Complete genomic molecule (chromosome, organelle genome, plasmid) | Reference |
| NG_ | Genomic region (for example a gene with flanks) | Reference |
| NT_ / NW_ | Genomic contig or scaffold | Assembly |
| NM_ | mRNA (protein-coding transcript) | Curated |
| NR_ | Non-coding RNA | Curated |
| NP_ | Protein | Curated |
| XM_ / XR_ / XP_ | mRNA / ncRNA / protein | Model, predicted by the annotation pipeline |
The pattern to remember: N = curated, X = predicted; the second letter gives the molecule (M mRNA, R ncRNA, P protein, C chromosome, G gene region).
Protein: UniProt
UniProt (run by EMBL-EBI, the Swiss Institute of Bioinformatics and the Protein Information Resource) has several parts:
| Component | What it is |
|---|---|
| UniProtKB/Swiss-Prot | Reviewed, manually annotated: function, domains, PTMs, variants, literature, evidence. Low redundancy |
| UniProtKB/TrEMBL | Unreviewed, automatically annotated translations of INSDC coding sequences; entries move to Swiss-Prot when curated |
| UniParc | Archive of every unique protein sequence, with its source records; no annotation |
| UniRef100 / 90 / 50 | Sequences clustered at 100%, 90% and 50% identity, to speed up searches |
| Proteomes | One proteome per sequenced organism, with reference proteomes marked |
Most protein sequences today are conceptual translations of genes, not direct protein sequencing, so their annotation depends on gene prediction being right (Phylogenetics, Comparative Genomics and Annotation).
Genome portals and organism databases
- Ensembl (EMBL-EBI with the Wellcome Sanger Institute), the UCSC Genome Browser, and NCBI’s Genome Data Viewer display assemblies with genes, transcripts, variants, conservation and regulatory tracks.
- NCBI Gene gives one page per gene, linking everything else.
- Model organism databases: FlyBase (Drosophila), WormBase (C. elegans), SGD (Saccharomyces), TAIR (Arabidopsis), MGI (mouse), ZFIN (zebrafish).
- Non-coding RNA: Rfam (RNA families with covariance models), miRBase (microRNAs), RNAcentral.
Variation and disease
| Database | Holds |
|---|---|
| dbSNP | Short variants (SNVs, small indels); IDs begin rs (reference SNP cluster) |
| dbVar / DGV | Structural variants and copy-number variants |
| ClinVar | Variant–condition assertions with clinical significance (pathogenic, likely pathogenic, VUS, likely benign, benign) |
| OMIM | Mendelian disorders and genes, curated text summaries |
| gnomAD | Population allele frequencies from large exome and genome collections |
| COSMIC | Somatic mutations in cancer |
Expression Databases
| Database | Host | Content |
|---|---|---|
| GEO (Gene Expression Omnibus) | NCBI | Functional genomics archive: microarray and sequencing-based expression, methylation, ChIP data. Record types: GPL (platform), GSM (sample), GSE (series), GDS (curated dataset) |
| ArrayExpress | EMBL-EBI | European counterpart of GEO (now served within EBI’s BioStudies) |
| Expression Atlas | EMBL-EBI | Re-analysed baseline (which tissue) and differential (which condition) expression |
| GTEx | — | Expression across many normal human tissues, linked with genotype (eQTLs) |
| Human Protein Atlas | — | Protein localisation in tissues and cells by antibody staining, plus RNA data |
| SRA | NCBI | Raw sequencing reads behind RNA-seq studies |
Reporting standards. Expression submissions follow minimum-information standards so others can reuse them: MIAME (Minimum Information About a Microarray Experiment) and MINSEQE for sequencing experiments. Both ask for raw data, processed data, sample annotation, experimental design and protocols.
Structure Databases
The Protein Data Bank
The PDB is the single global archive of experimentally determined 3D structures of proteins, nucleic acids and their complexes. It was founded in 1971 and is now managed by the worldwide PDB (wwPDB): RCSB PDB (USA), PDBe (EMBL-EBI), PDBj (Japan) and the BMRB (NMR data). Cryo-EM density maps go to the EMDB, with the fitted model in the PDB.
What a PDB entry carries:
- A four-character ID (a digit followed by three alphanumerics, for example 1ABC style).
- Experimental method: X-ray crystallography, NMR or electron microscopy.
- Atomic coordinates in
ATOM(standard residues) andHETATM(ligands, ions, water, modified residues) records. - Resolution in Å for X-ray and cryo-EM (smaller is better), R-factor and R-free for crystal structures (lower is better; R-free is computed on reflections left out of refinement, guarding against over-fitting), the B-factor (temperature factor: atomic displacement or disorder) and occupancy.
- Formats: the legacy fixed-column PDB format and mmCIF/PDBx, now the standard archive format (it handles very large structures that the old format cannot).
Related structure resources:
- AlphaFold Protein Structure Database (EMBL-EBI with Google DeepMind): predicted, not experimental, models with per-residue confidence (Protein Structure Prediction).
- NCBI MMDB (structures linked to Entrez), NDB (nucleic-acid structures).
- SCOP and CATH: hierarchical classifications of domains by structure (covered on the modelling page).
- Structure viewers: PyMOL, Chimera/ChimeraX, Jmol, RasMol, VMD (also used for dynamics trajectories).
Pattern, Motif and Family Databases
When a new protein has no close relative of known function, its conserved motifs and domains are the best clue. Different databases represent the “signature” of a family in different ways.
| Database | Signature type | Notes |
|---|---|---|
| PROSITE | Patterns (regular expressions) and profiles | Patterns are short, exact rules; profiles score whole domains |
| PRINTS | Fingerprints: groups of several conserved motifs | Diagnostic power from requiring several motifs together |
| BLOCKS | Ungapped multiple-alignment blocks | The BLOSUM matrices were derived from it |
| Pfam | Profile hidden Markov models (HMMER) of domain families | Seed alignment plus full alignment; clans group related families; now served through InterPro |
| SMART | HMMs of signalling and extracellular domains | Domain architecture diagrams |
| CDD (NCBI) | Position-specific scoring matrices of conserved domains | Searched by RPS-BLAST |
| InterPro | Integrates the member databases above (and others) into single entries | InterProScan runs them all on a query |
| TIGRFAMs, PANTHER, SUPERFAMILY | HMM families (function-level and structure-based) | Also InterPro members |
Reading a PROSITE pattern
| Symbol | Meaning |
|---|---|
x | Any amino acid |
[ST] | Any one of the listed residues (here Ser or Thr) |
{P} | Any residue except those listed (here not Pro) |
x(2) / x(2,4) | Exactly 2 / between 2 and 4 positions of any residue |
- | Separates positions |
< / > | Pattern anchored at the N- / C-terminus |
The N-glycosylation sequon is written N-{P}-[ST]-{P}: Asn, then any residue but Pro, then Ser or Thr, then any residue but Pro.
Ontologies, Pathways and Interactions
Gene Ontology (GO)
GO is a controlled vocabulary for describing gene products in three independent ontologies:
| Ontology | Question it answers | Example term |
|---|---|---|
| Molecular function | What does the product do at the molecular level? | protein kinase activity |
| Biological process | Which larger programme is it part of? | apoptotic process |
| Cellular component | Where does it act? | mitochondrial inner membrane |
Terms form a directed acyclic graph (a term can have several parents, through “is_a” and “part_of” relationships), not a simple tree. Each annotation carries an evidence code: experimental (IDA, IMP, IGI, IPI, IEP), computational (ISS and relatives), author or curator statements, and IEA (inferred from electronic annotation, not reviewed by a curator). GO enrichment analysis asks whether a gene list from an experiment holds more genes with a term than chance predicts (a hypergeometric or Fisher test, then a correction for testing many terms at once).
Pathways and interactions
- Pathways: KEGG (metabolic and signalling maps, KO orthology groups), Reactome (curated human reactions), BioCyc/MetaCyc (organism-specific metabolism; EcoCyc for E. coli), WikiPathways.
- Interactions: STRING (known and predicted functional associations with a combined confidence score), BioGRID and IntAct (curated physical and genetic interactions).
- Models: BioModels stores curated quantitative models in SBML (Systems Biology Markup Language).
- Chemistry: PubChem, ChEMBL, DrugBank and ZINC (see Computer-Aided Drug Design).
- Literature: PubMed (citations and abstracts), PubMed Central (full text).
Retrieval systems
- Entrez (NCBI) links Nucleotide, Protein, Gene, Structure, PubMed, Taxonomy and the rest; a record’s “related” links are precomputed neighbours.
- EBI Search plays the same role at EMBL-EBI; Ensembl’s BioMart gives bulk, table-style downloads.
- Programmatic access through E-utilities and REST APIs is how real pipelines query these databases.
- India’s national repository for life-science data is the Indian Biological Data Centre (IBDC), hosted at the Regional Centre for Biotechnology, Faridabad.
File Formats You Must Recognise
| Format | Looks like | Holds |
|---|---|---|
| FASTA | Header line starting >, then sequence lines | Sequence only |
| GenBank flat file | LOCUS, DEFINITION, ACCESSION, VERSION, FEATURES (CDS, gene, exon with locations), ORIGIN then numbered sequence, ending // | Sequence plus annotation |
| FASTQ | 4 lines per read: @id, sequence, +, quality string | Reads with per-base quality |
| SAM / BAM | Tab-separated alignment lines (BAM is the compressed binary form; CRAM compresses further against a reference) | Reads aligned to a reference |
| VCF | Header lines ##, then CHROM, POS, ID, REF, ALT, QUAL, FILTER, INFO, FORMAT, samples | Variants and genotypes |
| GFF3 / GTF | 9 tab-separated columns: seqid, source, type, start, end, score, strand, phase, attributes | Gene and feature annotation |
| BED | chrom, start, end (+ optional columns) | Intervals; 0-based start, end not included |
| PDB / mmCIF | ATOM/HETATM records or key–value tables | 3D coordinates |
Phred quality scores
Each base in FASTQ has a quality \(Q\) related to the probability \(P\) that the base call is wrong:
$$\boxed{Q = -10\log_{10}P}\qquad\Longleftrightarrow\qquad P = 10^{-Q/10}$$| Q | Error probability | Accuracy |
|---|---|---|
| 10 | 1 in 10 | 90% |
| 20 | 1 in 100 | 99% |
| 30 | 1 in 1,000 | 99.9% |
| 40 | 1 in 10,000 | 99.99% |
In the common Phred+33 encoding, the quality character’s ASCII code minus 33 gives Q. The character I has ASCII code 73, so Q = 40; + (43) gives Q = 10; 5 (53) gives Q = 20; ? (63) gives Q = 30.
Memory Tricks
Don’t Confuse
| This | vs | That |
|---|---|---|
| GenBank: primary, redundant, submitter-owned | vs | RefSeq: curated, non-redundant, NCBI-built |
| Swiss-Prot: reviewed | vs | TrEMBL: unreviewed, automatic |
| UniProtKB: annotated knowledgebase | vs | UniParc: sequence archive, no annotation |
| PDB: experimental structures | vs | AlphaFold DB: predicted models |
| GEO (NCBI) | vs | ArrayExpress (EBI): both expression archives |
| SRA: raw reads | vs | GenBank: assembled, annotated sequences |
| PROSITE pattern: exact rule | vs | Pfam HMM: probabilistic family model |
| PRINTS: fingerprints of several motifs | vs | BLOCKS: ungapped alignment blocks |
| FASTA format: sequence only | vs | FASTQ: sequence + per-base quality |
| BED: 0-based, half-open | vs | GFF/VCF: 1-based, inclusive |
| Resolution: lower number = better map | vs | B-factor: higher = more mobile or disordered atom |
Common Mistakes
Practice
All questions below are original practice questions written in CSIR NET style. They are not previous-year questions; this unit is new in the revised syllabus.
Level 1: Foundation
- Which section of UniProtKB contains manually reviewed entries?
- What kind of molecule does a RefSeq accession beginning with NP_ describe?
- Name the three GO ontologies.
Answers
- UniProtKB/Swiss-Prot.
- A curated protein sequence.
- Molecular function, biological process, cellular component.
Level 2: Part B Style
Q1 (practice). Match each database with the main type of data it archives.
| Database | Data | |
|---|---|---|
| P. GEO | 1. 3D coordinates from X-ray, NMR and cryo-EM | |
| Q. PDB | 2. Protein family signatures as regular expressions and profiles | |
| R. PROSITE | 3. Functional genomics and expression data | |
| S. DDBJ | 4. Primary nucleotide sequences |
- (a) P-3, Q-1, R-2, S-4
- (b) P-3, Q-2, R-1, S-4
- (c) P-4, Q-1, R-2, S-3
- (d) P-1, Q-3, R-4, S-2
Answer and explanation
(a). GEO is NCBI’s expression archive; PDB holds experimental 3D structures; PROSITE holds patterns and profiles; DDBJ is the Japanese INSDC partner for nucleotide sequences.
Q2 (practice). In a FASTQ file using Phred+33 encoding, a base has the quality character I (ASCII 73). The probability that this base call is wrong is:
- (a) 0.1
- (b) 0.01
- (c) 0.001
- (d) 0.0001
Answer and explanation
(d). Q = 73 − 33 = 40. \(P = 10^{-40/10} = 10^{-4}\).
Q3 (practice). Which statement about the PROSITE pattern C-x(2,4)-C-[LIVM]-{P} is correct?
- (a) The two cysteines are separated by exactly 2 residues.
- (b) The residue after the second cysteine must be Leu, Ile, Val or Met, and the next residue may be anything except Pro.
- (c)
{P}means the position must be Pro. - (d) The pattern must start at the N-terminus of the protein.
Answer and explanation
(b). x(2,4) allows 2 to 4 residues between the cysteines; [LIVM] means one of the listed residues; {P} excludes Pro. Anchoring to the N-terminus would need < at the start.
Q4 (practice). A BED file line reads chr2 99 200. In 1-based, inclusive coordinates, this interval covers:
- (a) bases 99–200, 102 bases
- (b) bases 100–200, 101 bases
- (c) bases 100–201, 102 bases
- (d) bases 99–199, 101 bases
Answer and explanation
(b). BED start is 0-based and the end is excluded: 0-based positions 99 to 199, which are 1-based positions 100 to 200. Length = end − start = 200 − 99 = 101 bases.
Level 3: Part C Style
Q1 (practice). A laboratory sequences a 3 Gb diploid genome with 150-base reads. Run 1 gives 200 million reads; run 2 gives a further 400 million reads. In run 1, the mean base quality across reads is Q30; in run 2 it is Q20.
(i) What is the mean coverage after run 1, and after both runs combined? (ii) Assuming Poisson coverage, what fraction of the genome is expected to be uncovered after run 1 alone? (iii) How many erroneous base calls are expected per 150-base read in each run?
Answer and worked solution
- Run 1: \(C = (150 \times 2\times10^8)/(3\times10^9) = 10\times\). Both runs: \(150 \times 6\times10^8 / 3\times10^9 = 30\times\).
- Uncovered fraction after run 1 = \(e^{-10} \approx 4.5\times10^{-5}\), about 0.005% of the genome (about 136,000 bases of 3 Gb).
- Expected errors per read = 150 × P. Run 1 (Q30, P = 10⁻³): 0.15 errors per read. Run 2 (Q20, P = 10⁻²): 1.5 errors per read, ten times more. Coverage helps against random errors (a consensus of many reads outvotes them) but low-quality runs inflate false variant calls at low-coverage sites, which is why reads are quality-trimmed first.
Q2 (practice). A student has a newly sequenced bacterial protein of unknown function. Consider the following plan:
(A) Search UniProtKB/Swiss-Prot first because its annotations are curated, then TrEMBL for more distant hits. (B) Run InterProScan to find domains, since it searches Pfam, PROSITE, PRINTS, SMART and other member databases together. (C) Look up the protein’s crystal structure in the AlphaFold Protein Structure Database. (D) Use GO terms transferred from a homologue, noting that IEA-coded annotations were not reviewed by a curator.
Which statements describe sound practice?
- (a) A, B and D only
- (b) A and C only
- (c) B, C and D only
- (d) A, B, C and D
Answer and worked solution
(a) A, B and D.
- (A) Sound: curated hits first, then the larger unreviewed set.
- (B) Sound: InterPro integrates the signature databases, and InterProScan runs them in one go.
- (C) Wrong: AlphaFold DB holds predicted models, not crystal structures. Experimental structures are in the PDB. A predicted model can still be useful, but it is not a crystal structure.
- (D) Sound: function transfer by homology is standard, and the evidence code tells you how much to trust it.
Q3 (practice). A GenBank record has ACCESSION AF000123 and VERSION AF000123.3. Its FEATURES table lists CDS join(101..250,401..601), and the CDS includes the stop codon. Which statements are correct?
(A) The sequence has been revised twice since its first release. (B) The coding sequence is 351 nucleotides long and encodes a protein of 116 amino acids. (C) Positions 251–400 are absent from the mature transcript, consistent with an intron. (D) The same record can be retrieved from ENA and DDBJ with the same accession.
- (a) A, C and D only
- (b) A, B, C and D
- (c) B and C only
- (d) C and D only
Answer and worked solution
(b) A, B, C and D.
- (A) Version .3 means two revisions after .1.
- (B) Exon lengths: 250 − 101 + 1 = 150 and 601 − 401 + 1 = 201; total 351 nt = 117 codons. One codon is the stop, so the protein has 116 amino acids.
- (C) The
joinskips 251–400 (150 nt), an intron in a genomic record. - (D) INSDC partners share accessions.
Part C habit: always check that a CDS length divides by 3. A remainder means a partial CDS (marked < or > in the location), a frameshift, or an annotation error.
Quick Revision Table
| Item | Fact |
|---|---|
| Primary vs secondary | Primary = raw, submitted, redundant (GenBank, PDB, GEO); secondary = derived, curated (RefSeq, Swiss-Prot, Pfam) |
| INSDC | GenBank (NCBI), ENA (EBI), DDBJ (NIG); shared accessions, daily exchange |
| Raw reads | SRA and its ENA/DDBJ counterparts |
| RefSeq | N = curated, X = predicted; NC chromosome, NM mRNA, NR ncRNA, NP protein |
| UniProtKB | Swiss-Prot reviewed; TrEMBL automatic; UniParc archive; UniRef clusters |
| Expression | GEO (GPL/GSM/GSE/GDS), ArrayExpress, Expression Atlas, GTEx; MIAME, MINSEQE |
| Structure | PDB via wwPDB (RCSB, PDBe, PDBj, BMRB); EMDB maps; mmCIF standard; AlphaFold DB predicted |
| Quality of a structure | Resolution (Å, lower better); R-free; B-factor = mobility |
| Patterns and families | PROSITE (regex + profiles), PRINTS (fingerprints), BLOCKS, Pfam (HMMs), SMART, CDD, InterPro (integrates) |
| PROSITE syntax | x any; [ ] allowed; { } forbidden; x(n,m) range; < > termini |
| GO | Function, process, component; DAG; evidence codes, IEA = electronic |
| Pathways / interactions | KEGG, Reactome, BioCyc / STRING, BioGRID, IntAct; SBML models in BioModels |
| Variation | dbSNP (rs IDs), ClinVar, OMIM, gnomAD, COSMIC |
| Formats | FASTA, GenBank flat file, FASTQ, SAM/BAM, VCF, GFF/GTF, BED (0-based), PDB/mmCIF |
| Phred | Q = −10 log₁₀ P; Q20 = 1%, Q30 = 0.1%; Phred+33 |
| Coverage | C = LN/G; uncovered fraction e^(−C) |
Key Takeaways
What’s Next
- Sequence Alignment, BLAST, FASTA and Multiple Alignment — how these databases are searched.
- Phylogenetics, Comparative Genomics and Gene Annotation — where the annotation in these records comes from.
- Experimental Design, Sampling and Data — Part A’s view of data quality and reproducibility.
- Chapter overview and Quick Revision.