Fig. 1: Schematic overview of the bioinformatics pipeline used for the prediction of allele-specific miRNA-binding sites in breast cancer risk variants identified in GWAS.

Genome-wide significant (\(P\le 5\times 1{0}^{-8}\)) SNPs associated with breast cancer risk in published GWAS were retrieved from the GWAS Catalog and from a GWAS meta-analysis.⁴⁹ Proxies in high linkage disequilibrium (\({r}^{2}\ge 0.8\)) were obtained through SNAP v2.2, using data from the pilot release of the 1000 Genomes Project for the CEU population. The biomaRt R package (v2.34.2) was used to retrieve genomic annotations from the Ensembl database v92. Risk-associated SNPs (raSNPs) were filtered for their location either in the 3′-UTR of protein-coding genes (PCGs) or in miRNA genes. Next, allele-specific miRNA-binding predictions were performed by modifying the input of two prediction algorithms—TargetScan and miRanda. First, each raSNP allele (reference and alternative) located at the 3′-UTR of PCGs was independently evaluated for putative miRNA binding through the algorithms. Then, allele-specific miRNA-binding predictions for each SNP were obtained by comparing each output file of corresponding SNP alleles and extracting their differences in miRNA binding. miRNA-binding predictions common to both algorithms were filtered for: (i) context++ score absolute difference (\(| \Delta\) context++ score\(|\)) \(\ge 0.151\); (ii) miRNA expression (\({\mathrm{RPM}}\,>1\)) in adjacent-normal breast tissue from the miRmine database and PCG expression (\({\mathrm{TPM}}\,>1\)) in normal breast from the GTEx Project; and (iii) evidence of PCG cis-regulation in normal breast tissue from in-house differential allelic expression analysis and eQTL data from the GTEx Project.