Gene function & annotation

Cross-Species Chromatin Marker Prediction: Linking Regulatory Activities and Complex Evolutionary Traits

---

Abstract

Changes in gene expression are thought to be a major cause of complex phenotypic diversity across vertebrates. However, no current model can reliably associate phenotypes across species through genome comparisons alone. Evolutionary conservation is seen with functional sequences, suggesting that conserved sequences within non-coding regions are more likely to code for regulatory elements. However, due to the large number of loci within non-coding regions relative to coding regions, validating genome regulatory annotations is costly and time consuming as seen from the work in the Functional Annotation of Animal Genomes (FAANG) project. To tackle this problem, we set out to accomplish two objectives. 1. Predict chromatin marks across mammalian species via convolutional neural network (CNN) classification. 2. Predicting phenotypes across species via phylogenetically aware linear regressions. To predict chromatin markers across mammals, we use Assay for Transposase-Accessible Chromatin (ATAC) and Chromatin Immunoprecipitation (ChIP) sequencing datasets to train neural networks to recognize orthologous sequence structure associated with each category of chromatin marker. Our work is based on Kaplow et al. 2022 and 2023, which used ATAC-seq data alone to successfully predict cross-species enhancers. We aim to extend the work to also include active histone modifications. Predictions are done for the following chromatin marker categories: ATAC (open chromatin), H3K27ac (active enhancers/promoters), H3K3me3 (active promoters). Current areas under the Receiver Operating Characteristic (ROC) curve range from 0.61 - 0.79 and areas under the Precision Recall (PRC) from 0.60 - 0.82 depending on the chromatin mark being predicted. We are currently focusing on muscle chromatin markers and body mass phenotypes across more than 200 species. We will perform a phenotype association for each chromatin mark's predictions. The phylogeny-aware linear regression takes phylogenetic branch lengths from the root and pairwise most recent common ancestor as their variance and covariance to take into account evolutionary time differences associated with each species. Analyses using methods including JWAS BayesC which gives estimated and true phenotype correlation accuracy of 0.6830 ± 0.132. Current next steps include CNN parameter optimization, training of CNNs for all histone mark prediction, addition of transformer attention layer and implementation of phylogenetic variance covariance. By having accurate chromatin marker predictions we can inform animal breeders about important potential regulatory genomic regions that they can focus on depending on their needs, and by observing patterns on chromatin marker heritability across mammals, we can identify conserved regulatory elements and how they have changed over evolutionary time.

Keywords: 2026

How to Cite:

Cheng, H., Kaplow, I. & Novoa, D., (2026) “Cross-Species Chromatin Marker Prediction: Linking Regulatory Activities and Complex Evolutionary Traits”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2287381. doi: https://doi.org/10.31274/wcgalp.24284

Rights: 1