The Gold Standard of Bioinformatics

A sequence, count matrix or image collection can support many questions. The choice of question determines which comparisons and methods are useful.

You might ask whether two sequences share a meaningful relationship, whether expression differs between groups, or whether a spatial pattern remains after accounting for sample structure. These are distinct analytical tasks.

Start by naming the biological entity, the comparison and the intended conclusion. Then identify the data needed to support that conclusion. A method that produces an attractive plot may still answer a different question.

EMBL-EBI’s introductory resources describe a field spanning storage, retrieval, comparison and modelling of biological information. The existing Bioinformatics mechanism article connects those tasks into a broader explanation.

A useful first exercise is to complete this sentence: “These observations would help us determine whether…”. Keep the wording specific enough that a result could change the answer.

The next step is to write down the possible outcomes before selecting software. That gives the analysis a purpose and makes it easier to recognise when an output does not resolve the original biological question.

Reference: What Is Bioinformatics? — EMBL-EBI Training.

Continue into: Main Bioinformatics mechanism.

Contents · Next chapter

A row labelled “sample 1” is difficult to interpret without knowing what it represents. It could describe a participant, tissue preparation, cultured population, individual cell or technical measurement.

Those distinctions affect the comparison. Repeated measurements from one preparation address different variation from independently collected biological material.

Metadata preserve the context needed to understand the records. Relevant fields may include sample identity, organism, collection conditions, preparation, measurement method and the relationship between repeated observations.

A useful record also distinguishes what was measured from what was inferred. A cell label assigned computationally has a different origin from a label established through another observation.

The Experimental Factor Ontology and related vocabulary guides help make descriptions comparable. Their value comes from preserving meaning across datasets.

For a learner, trace one plotted point back to the biological unit it represents. If that relationship is unclear, repair it before interpreting a difference between groups. An analysis becomes more informative when the unit of observation and the unit of inference remain connected.

Continue into: Experimental Factor Ontology.

Contents · Previous chapter · Next chapter

Names make records readable, while identifiers help connect them reliably. A gene symbol, transcript accession and protein accession can refer to different entities even when they concern the same biological system.

A mapping between identifiers needs its own context. Which organism, database release and entity type were used? Could several records map to one label, or one record map to several forms?

Versions can matter because sequences, annotations and reference resources are revised. A result may become difficult to reconstruct if its record of the reference is incomplete.

HGNC, NCBI Taxonomy, Protein Ontology and the guide to translating gene and protein identifiers provide focused routes into these distinctions.

For students, compare the readable name with the underlying entity. Is the claim about a gene, one transcript, a protein form or an annotation associated with it?

The useful next move is to preserve the original identifier alongside any converted label. That makes joins, comparisons and later updates easier to inspect without losing the meaning of the starting record.

Continue into: HGNC · NCBI Taxonomy · Protein Ontology · Gene and protein identifiers across languages.

Contents · Previous chapter · Next chapter

A reference sequence provides a framework for comparison. It has an identity, an organisation and, where applicable, an assembly version. Coordinates become meaningful in relation to that framework.

Two records can use the same chromosome label while referring to different assemblies. Their position numbers should therefore not be assumed to identify the same underlying sequence.

Assembly changes may involve more than a simple shift. Mapping coordinates between references can require a defined relationship and can leave some intervals unresolved or ambiguous.

The FASTA, refget, AGP and Chain Files articles supply different parts of this explanation: sequence content, reference identity, assembly organisation and coordinate mapping.

For learners, attach the reference identity to every coordinate statement. Then ask whether a comparison uses the same reference and compatible conventions.

This is a practical habit with a large payoff. Before proposing a biological reason for two records disagreeing, establish whether they were describing the same reference object in the first place.

Continue into: FASTA · refget · AGP · Chain Files.

Contents · Previous chapter · Next chapter

A file format specifies how information is represented. It can define fields, coordinate conventions, allowed values and relationships between records. Reading the numbers correctly requires those rules.

FASTA describes sequence records; FASTQ also carries per-base quality information. Alignment formats such as SAM and BAM describe relationships between reads and a reference. Variant records and feature annotations serve other purposes.

Coordinate conventions deserve particular attention. BED uses a zero-based start and an excluded end boundary. GFF3 describes feature coordinates using positive one-based positions, with the end position included for an ordinary interval.

For the same ordinary interval on the same reference, conversion therefore changes the start representation. Copying the same pair of numbers between the formats can change which bases are described.

The worked case applies that distinction with small fictional coordinates. The BED and GFF3 branches provide the focused explanations.

A useful student task is to identify the meaning of each field before calculating with it. A file that opens successfully can still be interpreted incorrectly if its conventions are missed.

Reference: Data File Formats — UCSC Genome Browser · GFF3 Specification — Sequence Ontology.

Continue into: BED · GFF3 · FASTQ.

Contents · Previous chapter · Next chapter

Biological measurements contain uncertainty, and different stages introduce different kinds of error. Sequence quality, sample preparation, contamination and missing information can affect an analysis in different ways.

A base quality score expresses information about a sequencing call under a particular scoring system. It is not a general guarantee that the entire sample, alignment or biological interpretation is correct.

Quality assessment should follow the question. A dataset suitable for one comparison may be insufficient for another. Filtering also changes the data being analysed and should have an explained purpose.

The FASTQ and DNA Sequencing guides connect sequence records with quality and error probabilities. The method articles on sequencing and measurement explain how evidence is generated before computational analysis begins.

For learners, ask which uncertainty a quality measure addresses. Then identify an important uncertainty it leaves unresolved.

The next step is to record consequential filtering and exclusion decisions. Another reader should be able to understand which observations contributed to the result and why those decisions fit the intended comparison.

Continue into: FASTQ · DNA sequencing and error probabilities.

Contents · Previous chapter · Next chapter

Sequence comparison can identify similar regions and support hypotheses about relationships. The interpretation depends on alignment length, coverage, composition, scoring, database context and other evidence.

A strong local match may concern only part of a molecule. The rest of the sequence can matter to a claim about overall function or evolutionary relationship.

BLAST’s expectation value relates a match score to the number of comparable chance matches expected in a search context. It is not the probability that a proposed biological function is correct. Database size and search settings belong in the account.

The NCBI documentation supplies the specialist reference. Alignment and indexing articles explain how computational records preserve the comparison.

For a learner, inspect what aligned and what remained unaligned. Then separate evidence of similarity from the more specific relationship being proposed.

A useful next question is what independent information would strengthen the interpretation: domain organisation, conserved features, known biology or an appropriate experimental result. The match starts the argument; the full claim needs its own support.

Reference: BLAST Frequently Asked Questions — NCBI.

Continue into: SAM · PSL and BLAT.

Contents · Previous chapter · Next chapter

Mapping asks how observations relate to a chosen reference. Assembly attempts to reconstruct longer sequence organisation from available evidence. Their outputs can support overlapping questions through different assumptions.

A mapped read is not automatically evidence for a unique location. Repeated regions and closely related sequences can create ambiguity. The record should preserve information needed to evaluate that assignment.

An assembly likewise needs evaluation. A long sequence representation is useful only when its joins, coverage and possible unresolved regions are understood.

SAM, BAM, CRAM, PAF and the assembly-related guides provide focused routes into these records. Indexes and compressed storage improve access while retaining their own requirements.

For students, explain the object being reconstructed or compared. Is the task to place a read, connect fragments, represent an assembled reference or compare two assemblies?

The next step is to identify where uncertainty enters the result. A more convenient file representation can preserve that uncertainty, but it cannot remove it merely by storing the output in a compact form.

Reference: High-throughput Sequencing Format Specifications — samtools.

Continue into: BAM · CRAM · PAF.

Contents · Previous chapter · Next chapter

A variant record represents a difference relative to a reference under specified conventions. Interpreting its biological consequence introduces further questions about evidence, molecular context and the relationship being claimed.

Transcript choice can influence a consequence description. A position may relate differently to different transcript models. Reference identity and allele representation also matter when comparing records.

VCF and BCF provide focused routes into variant data. Annotation and splicing-prediction articles address subsequent stages, each with its own assumptions and scope.

A computational prediction can help prioritise a question without establishing every biological effect. Its intended use and validation should remain visible.

For learners, follow a short chain: recorded difference, annotation, predicted consequence and observed effect. Mark which stages are supported by the available evidence.

The useful next move is to state the narrowest clear conclusion first. A well-described difference remains valuable even when its functional interpretation requires additional evidence. That clarity helps prevent a record label from becoming an unsupported causal story.

Continue into: VCF · BCF · ANNOVAR · SpliceAI.

Contents · Previous chapter · Next chapter

An ontology organises concepts and relationships so that records can be compared using shared meaning. It can help distinguish a molecular function, biological process, cellular location or other kind of description.

An annotation connects an entity with such a term. That connection has an evidential basis, which should remain available to the reader.

Gene Ontology provides a clear example: annotations carry evidence information, and different evidence types describe different sources of support. A computationally inferred association and a directly supported experimental association should remain distinguishable.

The Gene Ontology, Evidence and Conclusion Ontology, Sequence Ontology and EDAM articles provide focused routes into these ideas.

For a student, ask what the term means, which entity it describes and what supports the connection. A shared term improves comparability without making every associated claim equally established.

The next step is to preserve the reference and evidence alongside the annotation. That makes a vocabulary useful for scientific reasoning rather than merely a tidier set of labels.

Reference: Introduction to GO Annotations — Gene Ontology · Guide to GO Evidence Codes — Gene Ontology.

Continue into: Gene Ontology · Evidence and Conclusion Ontology · Sequence Ontology · EDAM.

Contents · Previous chapter · Next chapter

RNA sequencing commonly produces counts assigned to genes or other features. A larger count can reflect several influences, including biological abundance and aspects of measurement and processing.

Normalisation helps make a specified comparison meaningful. Its assumptions and the quantity being estimated matter. A normalised value is not automatically an absolute count of molecules in each cell.

DESeq2 and edgeR provide focused routes into models for replicated count data. Their interpretation begins with an appropriate design and data representation, rather than a choice of colourful plot.

For students, explain what a row and column in the count matrix represent. Then distinguish the raw observation, the adjusted comparison and the biological interpretation.

A useful next question is whether the compared samples differ in composition or another relevant property. Such differences can affect what an aggregate expression change means.

The aim is a comparison whose units, denominator and model remain clear. That makes the result easier to interpret and easier to check when the same data are viewed through another representation.

Continue into: DESeq2 · edgeR.

Contents · Previous chapter · Next chapter

Replicated observations help characterise variation, but their relationship to the biological question matters. Technical repeats and independent biological samples contribute different information.

A batch effect can arise when measurement or preparation conditions differ across groups of records. If every sample in one biological condition belongs to one batch and every sample in another condition belongs to another batch, the two explanations may be inseparable from those data.

Adding a batch label to software does not automatically solve that design problem. The official DESeq2 guidance explains how confounded design variables can prevent a model from estimating the intended effects.

The focused expression and single-cell comparison articles connect design with analysis.

For learners, draw a small table of biological group, preparation batch and independent sample. Look for comparisons that hold one factor steady while another changes.

The next useful move may be improved sampling or a narrower conclusion. A sophisticated analysis is most effective when the design supplies the information needed to distinguish the proposed biological effect from the relevant alternatives.

Reference: Analyzing RNA-seq Data with DESeq2 — Bioconductor.

Continue into: DESeq2 · muscat.

Contents · Previous chapter · Next chapter

Large datasets often involve thousands of comparisons. Even when each test is reasonable, interpreting the resulting collection requires attention to the number and structure of the tests.

An adjusted significance measure addresses a particular statistical procedure and its assumptions. It does not directly state the size of a biological effect or establish causation.

Effect size, uncertainty, experimental design and biological relevance remain part of the interpretation. A small estimated change can have strong statistical support in one setting while a larger change remains uncertain in another.

The DESeq2, edgeR, stageR and other focused method articles provide routes into different testing contexts.

For students, read the estimated effect alongside the uncertainty and the procedure used to assess it. Then ask which family of questions was tested and which decisions followed.

The next step is to resist turning a threshold into the whole explanation. A useful result states what changed, how strongly the data support that comparison and what further biological evidence would clarify its importance.

Continue into: stageR · edgeR.

Contents · Previous chapter · Next chapter

Single-cell methods can reveal variation concealed by an aggregate measurement. The resulting data still have a sampling design: cells may come from the same donor, preparation or experimental unit.

Many cells from one sample do not automatically provide the same evidence as many independent samples. The level of inference should match the design.

Cell labels and clusters also require interpretation. A computational group may reflect biological state, measurement conditions, analytical choices or several influences. Its boundaries are not established solely by a coloured embedding.

The Single-Cell RNA Sequencing guide explains the measurement route. muscat, dreamlet and related comparison guides address particular analytical settings and the role of sample structure.

For learners, trace one cell through its sample identity, quality information, assigned label and contribution to a comparison.

A useful next question is whether the conclusion concerns differences among cells in the observed material or differences between broader biological populations. Clear wording preserves the valuable detail while keeping the supporting evidence at the right scale.

Continue into: Single-Cell RNA Sequencing · muscat · dreamlet.

Contents · Previous chapter · Next chapter

Spatial methods connect molecular observations with positions, regions or tissue structure. Their units differ: one observation may represent a region containing several cells, while another has a different spatial resolution.

An inferred cell mixture is not the same kind of record as a directly measured cell position. Registration, segmentation and reference mapping can add further analytical steps.

Spatial Transcriptomics, RCTD, SPOTlight, DestVI and the surrounding guides supply focused routes into these distinctions. Other articles examine spatially varying genes, domains and relationships between neighbouring observations.

For students, identify what was measured at each location and which features were inferred from other data. Then ask how the spatial coordinates relate to the biological material.

Nearby observations can share properties for several reasons. A spatial association should not immediately become a claim about direct communication or a particular causal mechanism.

The next step is to choose the comparison that matches the question. Location makes the data richer when the measurement scale and inferential steps remain visible.

Continue into: Spatial Transcriptomics · RCTD · SPOTlight · DestVI.

Contents · Previous chapter · Next chapter

A network represents entities and relationships. In bioinformatics, an edge can mean correlation, a predicted regulatory influence, a curated interaction or another explicitly defined relationship.

Those meanings should remain distinguishable. A line inferred from expression data does not automatically establish direct physical contact or a causal effect.

Gene regulatory and cell communication methods use different evidence, assumptions and prior resources. The network directory connects guides on SCENIC, GENIE3, NicheNet, CellChat and many related approaches.

For learners, read an edge as a sentence. Which two entities are connected, what relationship is asserted and what evidence supports it?

Then ask what was absent from the analysis. A method can only propose relationships within the information and model available to it.

The next useful move is to identify a prediction that an independent observation could test. Networks become more informative when their edges have clear meanings and their proposed biological roles remain connected to evidence.

Continue into: SCENIC · GENIE3 · NicheNet · CellChat.

Contents · Previous chapter · Next chapter

Multi-omics analysis brings different measurements into a shared investigation. RNA, protein, chromatin and other records can contribute complementary information.

Integration begins with compatibility. Do the records describe the same samples, comparable populations or different material linked through an analytical model? Which identifiers and reference versions connect them?

The biological levels also differ. A change in RNA amount does not require an identical proportional change in protein abundance or activity. Agreement can support an explanation, while disagreement may reveal a useful question.

CITE-Seq, chromatin-accessibility methods and the proteomics format guides provide routes into specific measurements and representations. Systems Biology connects them with the behaviour of the larger system.

For students, draw the link between two datasets and label what justifies it. A shared gene name alone may leave important differences unresolved.

The next step is to explain what the combined analysis adds. Integration is most useful when it answers a clearer question than either dataset could address alone and preserves the origin of each contribution.

Continue into: CITE-Seq · Systems Biology · mzQuantML.

Contents · Previous chapter · Next chapter

A predictive model can learn patterns that help answer a biological question. Evaluating that ability requires a comparison that resembles the intended use.

If closely related records from the same participant or sample appear in both training and evaluation sets, apparent performance may overstate generalisation to new independent material.

The same concern applies when information from the full dataset influences preprocessing or feature selection before evaluation. The assessment should preserve the distinction between information available during model development and information used to test it.

Prediction and perturbation guides in the directory examine different modelling tasks. Their claims should be read alongside the data, benchmark and validation design.

For a learner, state what “new” means: a new cell, sample, donor, condition or population. Those are different tests.

A useful next move is to compare performance across meaningful subgroups and settings. A model becomes easier to judge when its successful use, uncertain boundaries and evidence for transfer are described concretely.

Continue into: scPRAM · trVAE · PhEval.

Contents · Previous chapter · Next chapter

A workflow connects inputs, transformations and outputs. Reconstructing an analysis can require reference resources, software versions, parameters, dependencies and the record of consequential decisions.

A command without its input identity may be difficult to interpret later. A result without a clear link to the command and reference may be equally difficult to check.

The GA4GH Tool Registry Service, EDAM, file-format and reference-identity guides address different parts of discoverability and comparability. A registry entry helps identify a tool; a reproducible analysis still needs its own complete context.

For students, trace one final table back through the transformations that produced it. At each stage, identify the input, method and output.

Then ask whether the original data can be distinguished from processed or edited versions. Preserving that relationship helps others understand which choices produced the result.

The next useful step is a concise record that another reader could follow. Documentation earns its value by making the analysis inspectable, repeatable and open to improvement.

Continue into: GA4GH Tool Registry Service · EDAM · refget.

Contents · Previous chapter · Next chapter

Bioinformatics often combines public resources with data collected under particular conditions. Availability and permitted use need to be understood for the actual dataset.

A public reference sequence and participant-linked genomic information can have different access arrangements. Removing a readable name does not automatically remove every possibility of linking a record with a person.

The practical task is to preserve the data-use context alongside the files. Which uses are permitted, which access conditions apply and what may be shared with the resulting analysis?

The htsget, refget, identifier and data-description guides help distinguish technical access from scientific meaning. A mechanism for transferring records does not itself answer every question about permitted reuse.

For learners working with teaching data, begin with a resource explicitly intended for that purpose and retain its source and terms.

The next step is to make sharing useful as well as appropriate: include clear descriptions, references and methods so that an authorised recipient can understand what the records and results represent.

Continue into: htsget · refget · DNA Profiling and Privacy.

Contents · Previous chapter · Next chapter

Methods are usually developed for particular data types and assumptions. Choosing among them begins with the intended comparison and the evidence available for that setting.

A benchmark is informative when its task, reference answers and evaluation measures resemble the proposed use. Performance on one dataset does not automatically settle suitability elsewhere.

Different methods may optimise different outcomes. Accuracy, calibration, sensitivity, computational requirements and interpretability can matter in different combinations.

The directory offers separate routes into expression analysis, splicing, spatial inference, networks and prediction. Their names are a starting point for choosing what to read, not a ranking of universal quality.

For students, write a short method comparison: question answered, required input, important assumption and evidence used to evaluate it.

The next useful move is to test whether a simpler approach already answers the question adequately. Complexity becomes valuable when it supplies a needed capability and the supporting evidence shows why it improves the intended task.

Continue into: PhEval · Mathematical Biology.

Contents · Previous chapter · Next chapter

A heatmap, embedding or genome track can make a complex dataset easier to inspect. Its interpretation depends on the quantities, transformations and visual encodings used.

A heatmap may show values scaled within each row. Its colours can therefore describe relative patterns without supporting a direct comparison of absolute abundance across rows.

An embedding also changes how distances are represented. A clear separation in a two-dimensional display should be interpreted alongside the method and original data structure.

Genome tracks add another set of conventions: reference identity, coordinate system, signal units and the relationship between intervals and annotations.

For learners, explain the colour scale, axes and unit represented by each point or row before writing a biological conclusion.

The next step is to choose a view that makes the relevant comparison visible. A useful figure helps a reader see both the pattern and the conditions needed to interpret it accurately.

Continue into: bedGraph · bigWig.

Contents · Previous chapter · Next chapter

Bioinformatics contributes to understanding genes, proteins, organisms, populations and environments. It also supports biotechnology, biological research and the development of further measurements.

A useful application preserves the bridge between a computational result and the decision it supports. A ranked candidate list, an inferred relationship and an experimentally established effect have different roles in that bridge.

Genetics, Molecular Biology, Biochemistry and Biotechnology provide the connected subject routes. Their explanations help interpret what a computational result could mean in the living setting.

For a reader, identify the scale of the claim. Does the evidence concern a sequence record, a measured sample, an organism or a wider population? What supports moving between those levels?

A clear statement can recognise a valuable finding while identifying the next observation needed.

The practical next move is to match the conclusion with its evidence. Bioinformatics becomes especially powerful when a computational insight helps formulate a biological question that can be checked through another appropriate source of information.

Continue into: Genetics · Molecular Biology · Biochemistry · Biotechnology.

Contents · Previous chapter · Next chapter

A student may learn a sequence of software actions while remaining uncertain about the biological question. Another may understand the biology but lose track of identifiers, units or file conventions.

Begin with a small task that exposes the first missing connection. Ask the learner to identify a matrix’s biological unit, explain a reference coordinate or distinguish an observed count from a model estimate.

Rebuild that connection using a small example whose records can be inspected directly. Then change the numbers or setting while preserving the underlying principle.

Teaching Biology supports the explanation of living systems. Mathematical Biology and the Education guide connect models, interpretation and learning that transfers to a changed task.

Progress is visible when the learner can explain the input, method, output and conditions in plain language, then recognise a mismatch in a new example.

The next exercise can be simple: compare two records that appear to disagree and determine whether the difference concerns biology, representation or the analysis. The worked case gives one concrete version of that task.

Continue into: Teaching Biology · Mathematical Biology · Education.

Contents · Previous chapter · Next chapter

Consider a fictional feature on a linear reference sequence called scaffold_A. Its ordinary inclusive coordinates in a GFF3 record are start 101 and end 110. A class converts the feature into a BED interval.

A student copies 101 and 110 directly into BED and assumes the feature has been preserved. The file opens, but the interval now describes nine bases under BED’s zero-based, end-excluded convention.

For this ordinary interval on the same reference, the correct BED start is 100 and the end remains 110. The GFF3 length is 110 − 101 + 1 = 10 bases. The BED length is 110 − 100 = 10 bases.

The corrected BED interval and the original GFF3 interval describe the same bases. The different start values express the different conventions.

This example assumes an ordinary non-wrapping interval, a fixed reference sequence and the same intended strand and feature identity. Other record types and special cases need their own specifications.

The class reports the reference, conventions and conversion, then checks both length and covered positions. That connects the whole subject: preserve identity, understand the representation and verify the transformation before proposing a biological explanation for a disagreement.

Reference: Data File Formats — UCSC Genome Browser · GFF3 Specification — Sequence Ontology.

Continue into: BED · GFF3 · Coordinate mapping with Chain Files.

Contents · Previous chapter · Explore the article directory