1.1 Two jobs that share a computer #
1.1.1 The definition these notes use #
Departments draw this line differently. Here, bioinformatics means making biological sense of sequence and structure data: storing, comparing, and annotating them, and deciding what a similarity allows you to say. Computational biology is the wider use of quantitative models of living systems, including population genetics, physical simulation, and pathway kinetics. Sequence work is one part of that wider field, not a rival to it.
The computer is not the distinction. A rate equation fitted to an enzyme assay is not bioinformatics. Reading a GenBank record is. A phylogenetic tree is both: the alignment is bioinformatics, and the substitution process is a model. The question decides. Which archived sequences resemble this one, under which score, in which database, on which date, is the first kind of question. What process generated the observations is the second.
1.1.2 What is being assumed #
A sequence here is a string of bases, with ambiguity codes, or of the twenty amino acids. The chemistry of those letters, and the reason a sequencer emits a call, belong to Biochemistry and Molecular Biology. These notes start once the call has been written down.
An alignment program does not know that tryptophan is bulky. It knows a score. If the matrix is wrong for the distance, the computation can be perfect and the biology useless. The later chapters are that warning in different clothes.
1.2 Why the archives exist #
1.2.1 From one finished read to a library of reads #
Archives are large because reading DNA became cheap, and because deposition is a condition of publication. A sequence with no accession is not in the shared record.
Three kinds of read still sit there, with different errors. Sanger sequencing is chain termination on one template, about 500–900 bases, accurate enough that a disagreement is usually biological or a cloning difference. Throughput was the limit. The chemistry is Molecular Biology. Older references are often one finished molecule, sometimes still carrying vector.
Short reads reversed the bargain: millions to billions of reads, often about 50–300 bases, commonly paired from the two ends of a fragment. Depth is what later separates a variant from an error. A read shorter than a repeat cannot name which copy it came from. The pair records a distance across an unsequenced middle. The middle itself is not data.
Long reads come from single molecules and run for thousands of bases, and on nanopore instruments much further. One molecule can cross a repeat, an isoform, or a junction. The historical price was the error rate, including homopolymer length. Consensus of many passes has narrowed that gap for some chemistries. Quote the instrument and the year. There is no timeless error percentage.
| Kind of read | What one read typically is | What it is good for |
|---|---|---|
| Sanger | One template, about 500–900 bases | Finished genes and older references |
| Short read | One of millions, often paired, about 50–300 bases | Depth: counts and small variants |
| Long read | One molecule, thousands of bases or more | Span: repeats, isoforms, junctions |
1.2.2 A reference is a coordinate system #
Later work is often reported against a reference assembly. The reference is not the ancestor, the healthy state, the common allele, or the organism in front of you. A variant is a disagreement with that chosen string. Change the reference and the variant list changes.
Deposition also freezes mistakes: contamination, a wrong species label, an open reading frame called through a frameshift. Derived databases then copy them. A record is a claim with a history.
1.3 Four kinds of biological object #
1.3.1 Genomes #
A genome data set is the DNA of a replicating thing, or an assembly offered as that DNA: chromosomes, plasmids, organelles, plus annotation. A “complete” genome is still a model. Eukaryotic assemblies break at centromeres, telomeres, and long tandem repeats unless the data were built to cross them. A reference chromosome is often a mosaic of donors, with haplotypes collapsed.
Genome Assembly
It can answer questions of content and order. It cannot say what a liver expresses on a Tuesday, which proteins are abundant, or what shape a chain adopts. Annotation that says otherwise has imported those claims or invented them.
1.3.2 Transcriptomes #
A transcriptome is the RNA of one sample: a cell type, a time, a treatment. There is no transcriptome of a species in the way there is, approximately, a genome. Poly(A) selection and ribosomal depletion do not count the same molecules.
Splicing makes several transcripts from one locus, so the data are molecules with abundances, not a gene list. RNA editing can make RNA disagree with the genome; that disagreement is not automatically a variant or an error. An RNA abundance is not a protein abundance and not a flux.
1.3.3 Proteomes #
A proteome is the proteins in a sample. It is not the translation of the genome. Splicing, alternative starts, cleavage, and modification produce chains the gene model does not spell. A database “proteome” is often only the set of translated gene models: useful, and not a measurement of what the cell contains.
Twenty letters with unequal frequencies cannot be scored as a bare match or mismatch without throwing away chemistry. That is why substitution matrices exist. Ask whether “the protein” means a gene model, a reviewed sequence, or a molecule someone measured.
1.3.4 Structures #
An archived structure is a coordinate model fitted to diffraction, nuclear magnetic resonance, or cryo-electron microscopy, under stated conditions: ligand, construct, and often one conformation of several. A predicted model has coordinates and no experiment. The shared file format does not make them the same evidence.
Structural Biology owns the path from sample to map, including what resolution means. Bioinformatics uses the deposited model as something to search and to doubt. The PDB entry is the boundary between those jobs.
1.4 Key terms, summary, and self-check #
Key terms. Bioinformatics; computational biology; Sanger read; short read; paired end; long read; reference sequence; genome; transcriptome; proteome; structure model.
Summary. Bioinformatics, in these notes, interprets sequence and structure data; computational biology is the wider quantitative modelling of living systems. Archives grew from finished Sanger reads, through deep short reads, to long reads. A genome is an assembly, a transcriptome is one RNA sample, a proteome is not a naive translation, and a structure is a model under stated conditions. A reference is a coordinate system, not an ancestor.
Self-check.
1. An enzyme assay is fitted with a rate equation and no sequence is used. Which of the two names applies?
2. Why can two papers list different “variants” for the same person?
3. Name one question a finished genome still cannot answer.
4. Why is a translated gene model not yet a proteome?
5. Why is “the long read is newer, so it is right” not a sufficient rule when it disagrees with a short read?
