1. What is Bioinformatics?

Bioinformatics is what happens when a biological question produces more data than a person can look at.

Learning objectives

  1. Describe the path from biological sample to analysable data.
  2. Name the major sequencing data formats and what each one holds.
  3. Explain why file size and file shape are different problems.
  4. Recognise that every dataset carries the fingerprints of how it was made.

A working definition

Bioinformatics is the use of computation to answer biological questions. That is deliberately broad, because the field is broad. In practice it usually means one of three things:

  • Processing — turning raw instrument output into something interpretable (reads → aligned sequences → variants).
  • Analysis — asking a question of the processed data (is this variant associated with disease?).
  • Interpretation — deciding what the answer means biologically, and whether to believe it.

Most training focuses on the middle step. Most mistakes happen in the first and last.

Where the data comes from

Every dataset in this workshop began as material from a person, a virus, or a culture. The chain looks like this:

sample → extraction → library prep → sequencing → basecalling → reads

Each arrow loses something and adds something. Extraction favours some templates over others. Library prep introduces PCR duplicates and adapter sequence. Basecalling makes probabilistic guesses at each position.

ImportantThe central point of this whole section

By the time data reaches your laptop, it has been transformed five times. Artifacts from any of those steps are indistinguishable from biology unless you know the steps happened.

The formats you will meet

Format Holds Notes
FASTA Sequences + names No quality information. Deceptively simple.
FASTQ Sequences + per-base quality The raw output of most sequencers.
SAM/BAM Reads aligned to a reference BAM is compressed SAM. Needs an index.
VCF Variants relative to a reference Positions are 1-based. Meaningless without knowing the reference.
Newick Tree topology + branch lengths A tree in a line of text.

A FASTA record:

>hbv_R02B35_consensus
GCTGCTATGCCTCATCTTCTTGTTGGTTCTTCTGGACTATCAAGGTATGTTGCCCGTTTG

A FASTQ record for the same read — same sequence, plus what the machine thought of each base:

@hbv_R02B35_read1
GCTGCTATGCCTCATCTTCTTGTTGGTTCTTCTGGACTATCAAGGTATGT
+
IIIIIIIIIIIIIIIIIIIIIHHHHHGGGGFFFEEDDCCBA@?>=<;:98

That fourth line is the part beginners discard and regret. Each character encodes a Phred score: the machine’s estimate of the probability this base is wrong.

NotePhred scores in one line

Q30 means a 1-in-1,000 chance the base is wrong. Q20 is 1-in-100. Q10 is 1-in-10. The scale is logarithmic, so the difference between Q30 and Q20 is much larger than it looks.

Why “big data” is the wrong frame

Sequencing data is large, but size is the least interesting problem. The hard parts are:

  • Shape — data arrives with samples in columns and features in rows, or the reverse, and nothing tells you which.
  • Missingness — a blank cell might mean “not measured”, “measured as zero”, or “measured and failed QC”. These demand different handling.
  • Provenance — the same sample sequenced twice on different runs will not produce identical results.

You will meet all three in the hands-on case study, where a dataset that looks clean turns out to have a fatal design problem that no amount of computation can fix.

Check yourself

  1. You are given a FASTA file and asked to assess sequencing quality. What do you tell the person who asked?
  2. A collaborator sends a VCF with no reference genome specified. What can you conclude from it?
  3. Why might the same sample yield different variant calls on two runs?

Answers are discussed in the exercise.