ChIP-seq

The primary analytical output is peaks—genomic regions with statistically significant enrichment of mapped reads over background. These peaks represent putative binding sites. Algorithms like MACS2 model the shift size of forward and reverse strand reads to identify these regions. The fold enrichment (signal over control) and false discovery rate (FDR) are critical metrics for filtering high-confidence binding events from noise.

A hallmark of high-quality ChIP-seq data is the bimodal distribution of reads on the positive and negative strands. Because sequencing occurs from the 5' ends of sheared fragments, reads pile up on the forward strand upstream of the binding site and on the reverse strand downstream. This shift distance correlates with the fragment length and is used to sharpen peak resolution.

ChIP-seq analysis is inherently comparative. A sonicated input DNA control (non-immunoprecipitated chromatin) is sequenced to model the background distribution of shearing bias, open chromatin, and copy number variations. True binding signals are identified by subtracting or normalizing against this control, preventing high-mappability or repetitive regions from being falsely called as peaks.

Data quality is quantified using metrics like the Normalized Strand Coefficient (NSC) and Relative Strand Correlation (RSC). NSC measures the ratio of fragment-length cross-correlation to background, while RSC compares fragment-length to read-length cross-correlation. High NSC (>1.05) and RSC (>0.8) values indicate strong enrichment and low background noise, essential for training robust predictive models.

The dynamic range of ChIP-seq is limited by sequencing depth. Saturation analysis plots the number of called peaks against the total mapped reads. As sequencing depth increases, peak discovery plateaus. For sharp marks like transcription factors, 10-20 million reads may suffice, but broad histone marks (e.g., H3K36me3) require significantly deeper sequencing to fully resolve large enriched domains.

The morphology of enrichment differs by target protein. Transcription factors typically produce sharp, punctate peaks (point-source). In contrast, certain histone modifications (e.g., H3K27me3, H3K9me3) form broad domains spanning kilobases to megabases. Peak-calling algorithms must be tuned specifically for these distinct spatial patterns to avoid fragmenting broad domains or merging discrete sharp peaks.

A high-throughput experimental technique, abbreviated as MPRA, that simultaneously tests the regulatory activity of thousands of synthesized DNA sequences. It provides functional validation for ChIP-seq-identified regulatory elements.

• Design: Synthesize candidate sequences with unique barcodes, clone into reporter plasmids
• Readout: Measure barcode abundance via RNA-seq to quantify transcriptional activity
• Scale: Test 10,000-100,000 sequences in a single experiment
• Integration: Used to confirm that ChIP-seq peaks correspond to functional enhancers rather than non-functional binding events

A model interpretability technique that attributes the prediction of a deep network to its input features by accumulating gradients along a path from a baseline to the actual input. Applied to ChIP-seq prediction models, it reveals the sequence motifs driving binding predictions.

• Mechanism: Computes the integral of gradients along a straight-line path from a neutral baseline (e.g., all zeros) to the input sequence
• Output: A nucleotide-resolution importance score for every position in the input DNA
• Validation: Recovers known transcription factor binding motifs without prior knowledge
• Advantage over attention: Provides true feature attribution rather than attention weights, which can be misleading