Genomics & Transcriptomics

The most complete free RNA-seq course. QC → trimming → alignment → quantification → DE analysis, with explanations of why each step exists.

Before any analysis: run FastQC on your raw reads and MultiQC to aggregate reports. Most analysis failures are detectable at this step.

Splicing-aware alignment is not just 'matching reads to a genome'. Understanding the splice junction problem explains why STAR exists and when it matters.

The de-facto standard for RNA-seq alignment. Fast, handles splicing, produces sorted BAM files. Know the key parameters before changing defaults.

The standard for bulk RNA-seq differential expression. Understand the negative binomial model and why you need count data (not FPKM) as input.

Non-negotiable. Inspect aligned reads at your top DE genes before writing a single result. Most analysis bugs are visible here.

Run it end-to-end at least once. It shows you what the steps are, in what order, with what defaults, and it handles the infrastructure so you focus on biology.

A practical walkthrough of integrating genomics data types for a real clinical question, good for seeing how the tools you're learning connect in practice.

How a specific genomic feature (repeat instability) becomes a clinical biomarker. A concrete example of genomics moving from methods to medicine.

What a reference genome is, how gene models are built, and why a genome version mismatch causes silent disasters downstream. More people have been burned by this than will admit.

Short-read Illumina vs. long-read Nanopore/PacBio, what the machine produces and how the read characteristics shape every downstream analysis choice.

PCA, clustering, multiple testing correction, and the EXACT statistical model behind DESeq2. If you're running these analyses, you need to understand what they're actually computing.

STAR is faster and more memory-hungry. HISAT2 is smaller and handles SNP-aware alignment. Know both, know when to choose each.

DESeq2 for most cases. edgeR when library sizes vary wildly. Read the vignettes, not just the wrapper tutorials: the model choices matter.

Seurat (R) and Scanpy (Python) are the two ecosystems. Pick one and learn it deeply: the concepts transfer, but don't context-switch until you've done a full analysis in one.

Run it. Read the MultiQC report it generates. Understand every module it runs. This is how real genomics production work is done.

Non-negotiable at every level. At researcher level: look at reads for every gene you intend to report.

Dr. Ömer Ali Bayraktar · 2023

How spatial transcriptomics extends RNA-seq by preserving tissue context. Cell2location, the Bayesian deconvolution method, is explained clearly by the author.

Building models that predict how regulatory elements drive gene expression in specific cell types, the frontier of gene regulation research.

Genomic instability, microsatellites, and why repeat regions matter in cancer genomics. Good for understanding how variant calling connects to clinical outcomes.

The paper behind the most widely used RNA-seq aligner. Read the algorithm section: it explains why STAR handles splicing the way it does, and why that matters for novel junctions.

The statistical model behind differential expression. Required reading before interpreting any DE result, know what 'shrinkage' means and why it's there.

Seurat v3 CCA-based integration. The method you'll use to merge datasets from different conditions or batches, understand it before relying on it.

Long reads solve problems short reads can't: full isoform resolution, structural variants, repeat regions. This review maps the analytical landscape: what tools exist and where they still fail.

CITE-seq, ATAC+RNA, spatial multi-omics: measuring multiple modalities per cell changes what questions you can ask. This review covers the current state of integration methods.

scATAC-seq analysis requires a different mental model than scRNA-seq. ArchR (R) and Signac (R, Seurat-integrated) are the dominant tools: both have excellent vignettes.

Deep generative models for integrating datasets across conditions, donors, and batches. More principled than CCA-based approaches for large, heterogeneous cohorts.

The standard aligner for long reads. Different parameter profiles for DNA vs. RNA (splice-aware), Nanopore vs. PacBio. Know the presets before tuning.

When nf-core doesn't do exactly what you need, you write your own pipeline. Snakemake's rule-based model is easier to reason about than Nextflow for one-off research workflows.

The frontier of regulatory genomics: models that predict gene expression from sequence features at cell-type resolution. Directly relevant to any work on enhancers, TF binding, or gene regulation.

Dr. Ömer Ali Bayraktar · 2023

Cell2location and spatial deconvolution in practice. If you're working with spatial data, hear from the group that developed the method.

Seurat v4's WNN framework for joint RNA+protein or RNA+ATAC analysis. The method you'll use for any multi-modal single-cell experiment.

The paper that introduced variational autoencoders to single-cell genomics. Read it to understand what scVI is actually doing when it 'integrates' datasets.

Bayesian deconvolution of spatial transcriptomics data. If you're working with Visium or similar, this is the method paper to understand.