8 Basic Bioinformatics Workflows

What makes a workflow reliable

A bioinformatics workflow is a documented series of steps that turns biological data into a result that can be checked and interpreted. A reliable workflow connects the question, the data source, the analysis method, and the biological conclusion.

Reproducibility means that someone else can identify the inputs and settings and repeat the analysis. It does not necessarily mean that every database search will return identical results: databases change over time.

Define the question and identify the data

Start with a question specific enough to guide the choices of data and method. Examples include identifying which known protein family is most similar to an amino-acid sequence, or finding which genes differ in expression between two conditions.

Choose a database suited to the question, then record its name, query terms or accessions, retrieval date, and relevant release or version information. GenBank and RefSeq provide nucleotide records, the Sequence Read Archive (SRA) stores sequencing data, and UniProtKB provides protein sequences and annotations. NCBI accession numbers identify records; an accession with a version suffix, such as NM_000000.2, specifies a particular sequence version.

Validate inputs and document methods

Check that the organism, sample, sequence type, and file format match the question. For FASTA, verify that every sequence begins with a > header and contains valid nucleotide or amino-acid symbols. For read data, retain sample identifiers and metadata that link each file to its biological condition.

Match the analysis tool to the question. Record the tool name and version, parameters, reference database or genome, and any filtering or preprocessing steps. Use the same documented steps for comparable samples.

Search for similar protein sequences

For an amino-acid sequence with unknown function, first check the FASTA header, unexpected characters, implausible length, and whether a nucleotide sequence may have been supplied by mistake as a protein sequence.

Use a protein-to-protein similarity search such as BLASTP against a database appropriate to the question. A curated protein set can make interpretation easier, while a broader set may find more candidates. Record the database and search settings. BLAST compares a query with database sequences and reports local alignments with statistical scores.

Evaluate matches and annotations

Evaluate candidate matches using several kinds of evidence. The E-value is the expected number of matches of that quality that would occur by chance in a database search; smaller values are generally more compelling. Also consider the alignment score, percentage identity, and, crucially, query coverage. A strong match spanning nearly the whole protein differs from a strong match limited to a short domain. Database size can affect the same E-value, so do not interpret that value alone.

Check the best-supported records for organism, protein name, domains, motifs, and annotation evidence. UniProtKB distinguishes reviewed from unreviewed entries. Experimental support and curated annotations provide useful context, but neither a high annotation score nor a top search result proves that the query has the same function.

Interpret protein evidence cautiously

State conclusions cautiously. For example, a sequence may be described as a strong candidate member of a protein family if it aligns across most of its length to several well-supported family members and shares the expected conserved domain. This evidence supports a functional hypothesis, not definitive proof.

Similarity alone does not establish identical function. Related sequences need not be equivalent in organism, biological role, or experimental context. Separate what the analysis directly reports from what you infer, and account for biological context when interpreting results.

Choose an alignment approach

When comparing a small number of sequences end to end, choose global alignment if the sequences are expected to correspond across their full lengths. Choose local alignment when looking for a matching region within sequences of different lengths.

Compare sequencing samples with a reference

For a public-data analysis, identify suitable sequencing runs and their sample metadata in the Sequence Read Archive (SRA), then retrieve the corresponding reads. Select a reference genome and annotation appropriate to the organism and study, and record the assembly accession and version. NCBI Datasets can provide genome sequences and related annotation files in a data package.

A typical analysis checks read quality, applies any documented filtering, aligns reads to the selected reference or uses a suitable alternative quantification method, and summarizes evidence at the level relevant to the question, such as genes or genomic regions. Comparisons between conditions must account for sample design and biological replicates.

Interpret reported differences in light of data quality, the reference and annotation used, and the limits of the statistical analysis. An observed association does not by itself establish a biological mechanism.

Inspect, save, and report results

Inspect data quality, alignment statistics, metadata, and biological context before interpreting results. Preserve original inputs, scripts or workflow definitions, intermediate results, logs, and final outputs. Report accessions, software versions, parameters, and enough methods detail for someone else to repeat the analysis.

Workflow platforms such as Galaxy retain tool settings and datasets in histories, from which reusable workflows can be created. Automating a workflow or saving it for reuse, while retaining the exact input data and settings, makes it easier to reproduce.

Check reproducibility before sharing

Before sharing results, check that the following details are recorded:

• Inputs: source databases, accessions, retrieval dates, sample metadata, and reference versions.

• Methods: software and version, parameters, filtering rules, and workflow order.

• Outputs: important intermediate files, final result tables, logs, and checks for errors or unexpected results.

• Interpretation: evidence supporting each conclusion, plausible alternatives, and limitations.

If a live database is queried again later, changes to the database may produce different results even when the same procedure is followed. A sound workflow begins with a clear question, uses suitable and identifiable data, applies a documented analysis, checks the results, and interprets them conservatively.

• Accession number