Concepts
1 Introduction
1.1 Who this is for
Two readers, and the book is arranged so neither has to wade through the other’s material.
The first is new to the field: comfortable with a terminal and some programming, but meeting sequence alignment, variant calling or single-cell analysis for the first time. For that reader, concepts come first and nothing is assumed. A term in bold is being defined right there.
The second has been doing this for years and wants to know what moved. The tooling landscape here turns over faster than any textbook cycle, and most of what is written down is either a decade stale or a single tool’s own documentation arguing for itself. For that reader, the implementation chapters and What changed are the entry points.
1.2 How the book is organized
Each subject is a part containing two chapters: concepts, then implementations. They are separate files, not two halves of one page, and that is deliberate.
Concepts age slowly. What a Burrows–Wheeler index does, why mapping quality is not alignment score, why spliced alignment is a different problem from contiguous alignment — none of that will be different in a decade.
Tool recommendations age in months. Keeping them in their own chapters means a revision touches only those files, each one carries its own review date, and a reader can see at a glance how fresh the advice is. It also means the stale-content risk is confined to the chapters that admit to it, rather than smeared across the whole book.
A teal badge under the chapter title marks a concepts chapter and a magenta one marks an implementations chapter. The same two colors carry the section rails and the tool cards throughout.
1.3 How to read a tool page
Every implementation chapter is generated from a machine-readable registry rather than written out by hand, which is what makes periodic revision tractable. Three things on those pages are worth knowing how to read.
The verdict. Recommended means start here. Viable means a defensible choice for a stated reason — usually a constraint like memory, or compatibility with something downstream. Legacy means it works but should not anchor a new pipeline. Avoid means actively discouraged, and the page will say why.
A verdict never appears without a reason. That is enforced by the registry’s schema, not by editorial discipline: a record marked recommended with no rationale will not validate, and the book will not build.
The upstream status. Separate from the verdict, and about the project rather than the advice: active means releasing, maintained means accepting fixes, dormant means no releases in about two years but still correct, superseded and deprecated mean do not start here. A tool can be dormant and still recommended — some problems are finished — but you should know which you are choosing.
The language chips. A solid chip is the language the tool is written in. An accented chip is a language you can drive it from through a binding, with the package name on hover. A language the tool does not support has no chip.
NoteWhere the gaps are reported
The gaps matter — a great deal of bioinformatics is C and C++ behind Python wrappers, R coverage is uneven and concentrated in Bioconductor, and Rust and Julia are strong in some corners and absent in others. That claim is made at the end of each implementation chapter, in prose, where it can be stated precisely and given its reasons.
It is deliberately not made by the chips. A chip row can only mark the four languages this book follows, so a struck-through chip would assert an absence it had never checked — silent about Java, Go and C++ — and for a tool that is a server or a command rather than a library, “no R binding” is not a gap but a category error. The chips say what a tool supports; the chapter says what the field lacks.
1.4 What this book does not do
It does not benchmark. Where a speed or accuracy claim appears it is attributed to a published comparison, not measured here — a benchmark that is not maintained is worse than none, and this book is committed to maintaining what it publishes.
It does not evaluate expensive code. Some pages show commands you would run on real data; those are illustrative, and the book does not claim to have executed them against a genome at build time.
It is not a protocol. Knowing that minimap2 -x map-ont is the current default does not tell you whether your experiment was designed correctly.