CentroFinder: a multi-feature framework for de novo prediction of fungal regional centromeres

Centromeres are essential chromosomal loci, yet their computational identification remains challenging due to rapid sequence evolution, high repeat content, and the absence of conserved defining motifs. This challenge is particularly pronounced in fungi, where centromere architectures vary widely in size, sequence composition, and chromatin organization, limiting the effectiveness of single-feature or motif-based prediction approaches.

We present CentroFinder, a fungal-specific computational framework for de novo centromere prediction from long-read sequencing-based genome assemblies. CentroFinder integrates multiple genomic and long-read-derived features into a weighted scoring model to identify loci where centromere-associated signals converge. Benchmarking against experimentally mapped centromeres in Cryptococcus deuterogattii, Magnaporthe oryzae, and Neurospora crassa showed that 27 of 28 predicted intervals overlapped the corresponding experimental domains, yielding 96.4% chromosome-level detection sensitivity. Application to 11 additional fungal genomes produced one contiguous predicted centromeric region per chromosome, supporting the transferability of the workflow for chromosome-level centromere prediction.

CentroFinder is freely available as open-source software at https://github.com/RahnamaLab/CentroFinder. The pipeline is designed for high-performance computing environments and leverages features derived from long-read sequencing data.