https://www.pierre-francois-roux.com/categories/genomics-tips/ Recent content in Genomics tips on Pierre-François Roux Hugo -- gohugo.io en-us © 2021 Pierre-François Roux Fri, 19 Jun 2026 09:50:00 +0000 https://www.pierre-francois-roux.com/2026/06/19/rank-transcription-factors-by-evidence-not-only-by-enrichment/ Fri, 19 Jun 2026 09:50:00 +0000 https://www.pierre-francois-roux.com/2026/06/19/rank-transcription-factors-by-evidence-not-only-by-enrichment/ Motif enrichment is a good starting point, but it is a weak ending point. A motif can be enriched because many related transcription factors share the same binding preference, because the region set is GC-rich, or because accessibility has changed without direct TF binding. When I try to prioritize candidate regulators, I prefer to build an evidence table. Each transcription factor gets several independent scores, and the interesting candidates are those supported by multiple layers. https://www.pierre-francois-roux.com/2026/06/19/make-enhancer-analyses-gene-aware-but-not-gene-blind/ Fri, 19 Jun 2026 09:40:00 +0000 https://www.pierre-francois-roux.com/2026/06/19/make-enhancer-analyses-gene-aware-but-not-gene-blind/ Enhancer analyses often fall into two traps. The first is to treat enhancers as isolated genomic intervals. The second is to assign every enhancer to the nearest gene and pretend the problem is solved. A better compromise is to make enhancer analyses gene-aware, but not gene-blind. Classify Enhancers Before Linking Them Start by defining interpretable enhancer classes. For example: Active enhancers: H3K27ac and H3K4me1 positive. Poised enhancers: H3K4me1 positive, low H3K27ac. https://www.pierre-francois-roux.com/2026/06/19/atac-seq-is-more-than-a-peak-list/ Fri, 19 Jun 2026 09:30:00 +0000 https://www.pierre-francois-roux.com/2026/06/19/atac-seq-is-more-than-a-peak-list/ Many ATAC-seq analyses stop at peak calling: open regions in condition A, open regions in condition B, differential accessibility. That is useful, but it leaves a lot of information on the table. ATAC-seq libraries carry several layers of signal: chromatin accessibility, fragment architecture, nucleosome positioning, transcription factor footprints, and sometimes even useful genetic information. Look At Fragment Classes The fragment length distribution is one of the first things I inspect. https://www.pierre-francois-roux.com/2026/06/19/use-pseudoreplicates-to-make-peak-sets-less-fragile/ Fri, 19 Jun 2026 09:20:00 +0000 https://www.pierre-francois-roux.com/2026/06/19/use-pseudoreplicates-to-make-peak-sets-less-fragile/ Peak sets are surprisingly fragile. Change one replicate, one depth threshold or one peak-calling parameter, and thousands of regions may appear or disappear. A useful way to reduce this fragility is to ask whether peaks are reproducible under resampling. This is the logic behind pseudoreplicates and IDR-style workflows. The Core Idea For each biological replicate: Randomly split aligned reads into two pseudoreplicates. Call peaks on the original replicate. Call peaks on each pseudoreplicate. https://www.pierre-francois-roux.com/2026/06/19/a-qc-checklist-before-trusting-chip-seq-or-atac-seq-peaks/ Fri, 19 Jun 2026 09:10:00 +0000 https://www.pierre-francois-roux.com/2026/06/19/a-qc-checklist-before-trusting-chip-seq-or-atac-seq-peaks/ Peak calling is not the beginning of an epigenomic analysis. It is a consequence of many upstream choices: library quality, mapping behavior, duplication, background, blacklisted regions and replicate structure. Before interpreting peaks, I like to ask a simple question: would I trust this experiment if I ignored the peak caller entirely? Start With The Library Check these before any biological interpretation: Per-base quality and adapter content. Fraction of aligned reads. https://www.pierre-francois-roux.com/2026/06/19/think-in-trajectories-not-only-in-contrasts/ Fri, 19 Jun 2026 09:00:00 +0000 https://www.pierre-francois-roux.com/2026/06/19/think-in-trajectories-not-only-in-contrasts/ Differential expression is often framed as a list of pairwise contrasts: treated versus control, late versus early, condition A versus condition B. That is useful, but it can flatten a time-course experiment into disconnected snapshots. When samples are ordered in time, a better first question is: which genes follow similar trajectories? A transient pulse, a delayed induction and a progressive repression can all have the same fold change in one contrast, but they usually mean different biology.