Please run the relevant commands in ../download_data.sh (most have been commented out to avoid downloading unnecessary data) in order to get the data files necessary for variant effect prediction.

Because Hedgehog only predicts methylation level changes at CpG loci, variant effect prediction entails computing the difference between methylation level predictions for a centered CpG locus given an off-center mutation. Given Hedgehog's input sequence length of 2048bp, variants can be up to 1024bp away from the center CpG locus. We only provide functionality for handling single-base mutations, not insertions or deletions.

Also note that Hedgehog excludes chrX and chrY from training to remove sex chromosome-specific methylation effects, so we recommend removing chrX and chrY variants from your variant set of interest as well.

As a result, the steps to run variant effect prediction are as follows:

• Step 0: If the CpG locus associated with the variant is known (e.g. mQTLs associate a variant to a CpG locus), skip this step. Otherwise, a variant could impact any CpG locus within 1024bp away. In this case, we recommend intersecting your variant file with the appropriate BED file of CpG locus +/- 1kb regions in ./data (we provide hg19 and hg38 files). Follow the steps in ./example to go from a test.vcf file to a BED file as described.

• Step 1: For every variant-CpG locus pair, create reference and alternative allele 1024bp sequences (centered at the CpG locus) and output each of these into reference and alternative allele sequence FASTA files. See ./example for an example script to do so (./example/cpg_variants_to_fasta.py) Additionally, we will soon provide notebooks associated with the manuscript results for mQTL analysis and HGMD pathogenic mutation prediction which can be adapted for your own analysis.

• Step 2: Run

sh fasta.sh <path-to-reference-fasta>/refs.fasta <output-directory>
sh fasta.sh <path-to-alternative-fasta>/alts.fasta <output-directory>

• Step 3: In our manuscript, we restrict our analysis to the CpG locus that is maximally impacted by a given variant by taking the maximum absolute difference score between the alternative allele predictions and the reference allele predictions, across all 296 methylation profiles predicted by Hedgehog.

• Step 4: To add additional context to your analysis, the methylation sequence class assignment of the CpG loci are available at ./data/berry.liftover_hg19.intersect_meseq_clusters.bed and ./data/berry.hg38.intersect_meseq_clusters.bed (depending on which genome version your variant coordinates use). Simply filter the BED file to the CpG loci that are impacted by your variant set and use the methylation sequence class assignment to get a high-level characterization of these loci based on methylation regulation patterns.