Machine learning and multi‐omic analysis reveal contrasting recombination landscape of A and C subgenomes of winter oilseed rape

TL;DR

This study uses machine learning and multi-omic data to show how recombination varies across the A and C subgenomes of rapeseed, revealing patterns linked to chromatin features.

Contribution

The novel use of machine learning to predict recombination hotspots in Brassica napus, revealing contrasting subgenome-specific recombination landscapes.

Findings

Recombination is suppressed in methylated, repeat-rich regions and enriched in gene-dense, transcriptionally active domains.

Random forest machine learning models achieved high accuracy in predicting recombination rates and hotspot locations.

The A subgenome shows clustered crossovers near subtelomeres, while the C subgenome has more evenly distributed recombination.

Abstract

Meiotic recombination is essential for generating genetic diversity, driving plant evolution, and enabling crop improvement, yet its uneven distribution across genomes constrains breeding efforts. Here, we investigated the multi‐omic landmarks that shape the recombination landscape in Brassica napus by integrating epigenomic, genomic, and transcriptomic data with recombination maps derived from large multiparental rapeseed populations. Predictive machine learning accurately predicted recombination rates and hotspot location using only feature information. Recombination was generally suppressed in centromeres and other repeat‐rich, methylated regions and enriched in gene‐dense, transcriptionally active domains. Proxies for chromatin configuration—such as DNA methylation, transposable elements, or genes—consistently achieved the highest predictive power with the random forest algorithm. We discovered distinct recombination landscape patterns between subgenomes, with crossovers clustering near subtelomeres in the A subgenome and more evenly spread across the C subgenome. Models trained on A subgenome data outperformed those based on the C subgenome, although combining both subgenomes improved overall accuracy.

We generated recombination maps in Brassica napus from a large set of crossovers detected across 5000 meiotic events from two large multiparental populations genotyped with a 15K single‐nucleotide polymorphism chip.Regarding methylation in the CHH (where H represents any nucleotide except guanine) context, we observed positive associations with recombination when restricting the analysis to transposable element (TE) bodies, reflecting the activity of CHH island silencing TEs near genes.We applied a machine learning approach on integrated multi‐omics and recombination data, achieving high accuracy in predicting recombination rate (overall R2 = 0.477) and hotspot location (overall area under the receiver operating characteristic curve = 0.823). Random forest emerged as the most robust algorithm for estimating feature importance amid multicollinearity.We confirmed the association of well‐known chromatin state indicators—mainly DNA methylation, TEs, genes, and gene expression—in shaping recombination in B. napus.We revealed distinct recombination landscapes between the A and C subgenomes: recombination was concentrated in the subtelomeric regions of the A subgenome, whereas it was more evenly distributed along the arms of the C subgenome. These patterns aligned with differences in the distribution of chromatin state markers between the two subgenomes.

We generated recombination maps in Brassica napus from a large set of crossovers detected across 5000 meiotic events from two large multiparental populations genotyped with a 15K single‐nucleotide polymorphism chip.

Regarding methylation in the CHH (where H represents any nucleotide except guanine) context, we observed positive associations with recombination when restricting the analysis to transposable element (TE) bodies, reflecting the activity of CHH island silencing TEs near genes.

We applied a machine learning approach on integrated multi‐omics and recombination data, achieving high accuracy in predicting recombination rate (overall R2 = 0.477) and hotspot location (overall area under the receiver operating characteristic curve = 0.823). Random forest emerged as the most robust algorithm for estimating feature importance amid multicollinearity.

We confirmed the association of well‐known chromatin state indicators—mainly DNA methylation, TEs, genes, and gene expression—in shaping recombination in B. napus.

We revealed distinct recombination landscapes between the A and C subgenomes: recombination was concentrated in the subtelomeric regions of the A subgenome, whereas it was more evenly distributed along the arms of the C subgenome. These patterns aligned with differences in the distribution of chromatin state markers between the two subgenomes.

Meiotic recombination creates genetic diversity, which is key for plant evolution and crop breeding, but it does not occur evenly across the genome. We studied this process in rapeseed (Brassica napus) by combining genetic maps with data on DNA, chromatin, and gene activity. Using machine learning, we could accurately predict where recombination happens. It was rare in repeat‐rich, methylated regions like centromeres but common in active, gene‐rich areas. We also found differences between the two subgenomes: crossovers clustered near chromosome ends in the A subgenome, while they were more evenly distributed in the C subgenome.

Figures