Single-Cell RNA-seq with BioLang

A tissue is not one thing. A blood sample contains many immune populations. A tumor contains malignant cells, immune cells, blood-vessel cells, connective tissue cells, and damaged cells. Bulk RNA sequencing mixes their RNA into one average. Single-cell RNA sequencing (scRNA-seq) measures thousands of those cells separately.

That extra detail is useful, but it creates a new problem: a colorful map can be easy to produce and easy to overinterpret. This book teaches both the workflow and the reasoning that makes the workflow defensible.

Who this book is for

• Biologists can connect laboratory decisions to computational artifacts.
• Researchers can turn a biological question into a reproducible analysis.
• Programmers can understand the data model, algorithms, and performance boundaries without first becoming molecular biologists.
• Clinicians can understand what a single-cell result does and does not say about a patient or treatment.
• Students and interested readers can begin without prior RNA-seq knowledge.

Each chapter answers five questions:

1. What is this concept?
2. Why does it matter?
3. How is it represented or analyzed?
4. When and where should it be used?
5. What can fool us?

What you will build

You will analyze a deterministic, tumor-like 10x count matrix with four known cell populations. The fixture is intentionally small enough for a laptop and is generated locally rather than stored in the repository. You will:

• inspect raw counts and quality metrics;
• remove implausible cells and rarely detected genes;
• normalize counts and select variable genes;
• compute PCA, a neighbor graph, Leiden clusters, and UMAP;
• inspect markers and propose cell-type labels;
• compare BioLang clusters with Scanpy and Seurat;
• plan a replicate-aware multi-sample analysis;
• record the decisions needed to reproduce the result.

A necessary boundary

This book teaches research analysis. A cluster, marker, or association is not a clinical diagnosis. Clinical use requires an independently validated assay, quality system, predefined decision rule, appropriate population, regulatory review, and human oversight.

BioLang’s singlecell package is useful for transparent preprocessing, exploration, and reproducible workflows. It does not make biological judgment automatic. The analyst remains responsible for study design, metadata, cell annotations, statistical units, and interpretation.

How to read it

Read Parts I and II in order on a first pass. Part III is organized by scientific question. Part IV should be read before publishing or handing a result to a collaborator. Terms in bold are collected in the glossary, and every package call is summarized in the API appendix.