Allelic variation is a critical determinant of agronomic traits in heterozygous crops. Most existing approaches define variation as reference-anchored differences, such as SNPs or structural variants, confining allelic diversity to variant feature coordinates. Here, we present AlleleMiner, a Python-based pipeline that phases diploid gene sequences directly from PacBio HiFi reads. Rather than relying on reference-based coordinate systems for allele representation, AlleleMiner uses the reference genome solely to identify target gene region sequences and performs de novo assembly of read sets at each locus, minimizing reference dependence and reconstructing phased allele sequences. Across 18 citrus cultivars, the pipeline achieved an average phasing output of 91.5% of 1,409 single-copy genes, coverage achieving. Coverage analyses using both real and simulated datasets indicated that a ∼30× HiFi depth is preferable for the stable recovery of heterozygous alleles, reducing potential allele dropout. Validation using pedigree information showed allele transmission patterns with known relationships. Using simulated haplotype data and the Citrus clementina assembly v1.0, AlleleMiner achieved complete-match reconstruction for both alleles at approximately 70% of loci. By enabling reference-minimized gene-level allele discovery, AlleleMiner provides a scalable framework for constructing allele databases and advancing marker-assisted and genomic selection in complex crops.
Kiryu et al. (Sun,) studied this question.