Cross-species plant genomes modeling at single nucleotide resolution using a pre-trained DNA language model

Zhai, J.; Gokaslan, A.; Schiff, Y.; Berthel, A.; Liu, Z.-Y.; Miller, Z. R.; Scheben, A.; Stitzer, M. C.; Romay, C.; Buckler, E. S.; Kuleshov, V.

doi:10.1101/2024.06.04.596709

bioRxiv · 10.1101/2024.06.04.596709

Cross-species plant genomes modeling at single nucleotide resolution using a pre-trained DNA language model

Zhai, J.·Gokaslan, A.·Schiff, Y.·Berthel, A.·Liu, Z.-Y.·Miller, Z. R.·Scheben, A.·Stitzer, M. C.·Romay, C.·Buckler, E. S.·Kuleshov, V.

View original source ↗Export metadata ↓

Abstract

Interpreting function and fitness effects in diverse plant genomes requires transferable models. Language models (LMs) pre-trained on large-scale biological sequences can learn evolutionary conservation and offer cross-species prediction better than supervised models through fine-tuning limited labeled data. We introduce PlantCaduceus, a plant DNA LM based on the Caduceus and Mamba architectures, pre-trained on a curated dataset of 16 Angiosperm genomes. Fine-tuning PlantCaduceus on limited labeled Arabidopsis data for four tasks, including predicting translation initiation/termination sites and splice donor and acceptor sites, demonstrated high transferability to 160 million year diverged maize, outperforming the best existing DNA LM by 1.45 to 7.23-fold. PlantCaduceus is competitive to state-of-the-art protein LMs in terms of deleterious mutation identification, and is threefold better than PhyloP. Additionally, PlantCaduceus successfully identifies well-known causal variants in both Arabidopsis and maize. Overall, PlantCaduceus is a versatile DNA LM that can accelerate plant genomics and crop breeding applications.

Explore related subjects

bioinformatics Preprint

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Zhai, J., Gokaslan, A., Schiff, Y., Berthel, A., Liu, Z.-Y., Miller, Z. R., Scheben, A., Stitzer, M. C., Romay, C., Buckler, E. S., Kuleshov, V.. 2024-06-05. Cross-species plant genomes modeling at single nucleotide resolution using a pre-trained DNA language model. https://doi.org/10.1101/2024.06.04.596709

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2026

Heterogeneous Graph Contrastive Learning for Drug-Gene-Disease Motif Prediction

Drug repurposing and target discovery offer critical strategies for advancing therapeutic development by uncovering the potential biological pathways and novel associations among drugs, genes, and diseases. However, experimental discovery remains expensive and time-consuming, which limits the scalability of large-scale studies. In addition, existing computational approaches often struggle to effectively integrate heterogeneous biomedical data, capture the complex higher-order topological signatures of biological interactomes, and generalize to unseen entities. Here, we present HANAMI (Heterogeneous grAph coNtrastive leArning for drug-gene-disease Motif predIction), a multi-view deep graph learning framework designed to model complex interactions among drugs, genes, and diseases. HANAMI integrates diverse heterogeneous biomedical knowledge, including chemical structures, genomic sequences, and clinical phenotypes, and leverages relation-aware topology encoding, structure-aware aggregation, and contrastive learning to enable accurate motif prediction with biological context from the network. Systematic evaluation on benchmark datasets shows that HANAMI achieves up to 6% improvements over existing state-of-the-art methods in predicting drug-gene-disease motifs. The framework further demonstrates strong inductive generalization, maintaining an [~]18% performance advantage in zero-shot settings involving previously unseen entities. Beyond predictive performance, HANAMI effectively prioritizes drug-disease relationships investigated in Phase II or III trials while identifying candidate genes that suggest plausible mechanistic links. Together, HANAMI provides a computational framework for interpreting complex biomedical interactions, offering a scalable foundation to accelerate drug repurposing and therapeutic innovation.

bioinformatics↗

bioRxiv2026

PTMExplorer: A Multi-Dimensional Integrative Visualization Platform for Protein Post-Translational Modification Function and Structure

Deciphering the functions of post-translational modifications (PTMs) is a critical bridge connecting large-scale modification proteomics data to mechanistic studies. However, most existing tools for visualizing PTM omics data are limited to site catalogs or single-dimensional feature displays. They lack the capability to simultaneously map user-derived differential modification sites onto multi-dimensional contexts, including protein three-dimensional (3D) structure, evolutionary conservation, functional sites, and disease associations. This limitation makes it difficult for researchers to rapidly assess the biological importance of candidate sites from among a vast number of differentially modified sites. Here, we present PTMExplorer, an interactive platform for the multi-dimensional visualization of protein PTMs. PTMExplorer comprises three core modules: PTM Inspector, built upon ProtVista, provides a multi-track, sequence-feature integrated view incorporating intrinsically disordered region (IDR) prediction (via flDPnn), surface accessibility calculation (via FreeSASA), and UniProt functional annotations; PTM 3D Locator, leveraging the Nightingale/Mol* engine, anchors modification sites onto AlphaFold/Protein Data Bank (PDB) 3D structures through residue mapping via PDBe-SIFTS; and PTM Overview, utilizing the R circlize package, presents a panoramic polar circos plot illustrating modification distribution and inter-group differential regulation. Additionally, three major disease-associated modification databases (PTMD, qPTM, and PhosCancer) are integrated as PTM-Disease Nexus, enabling co-localization comparison between user-defined differential sites and reported disease-related sites. PTMExplorer currently supports eight model organisms, accepts user-uploaded differential analysis results, and provides multi-dimensional annotations and various visualization options (https://www.bioladder.cn/PTMExplorer/). Using a multi-omics dataset from hepatocellular carcinoma (18 patients, 9 modification types) as a case study, we demonstrate the practical utility of PTMExplorer in screening potential biomarkers, revealing multi-modification coordination mechanisms, and distinguishing between absolute and relative quantification patterns.

bioinformatics↗

bioRxiv2026

Integrative analysis of the MDM2 promoter switch and cellular lineage plasticity in colorectal cancer: a contrast between the autonomous-proliferation type (CIN/CMS2) and the environment-adaptive type (MSI/gastric metaplasia)

Background: Biomarkers that stratify colorectal cancer (CRC) by therapeutic responsiveness and are measurable directly in biopsy specimens remain insufficiently established. We investigated whether usage of the dual MDM2 promoters (P1/P2) acts as a molecular switch separating two diametrically opposed CRC phenotypes: a chromosomal instability type and an environment adaptive type (microsatellite instability/serrated pathway with gastric metaplasia). Methods: Sixty three organoid samples from 22 patients with CRC were classified morphologically by deep learning (VGG16) and molecularly by an MDM2 Splicing Index derived from expression arrays. The P1 and P2 signatures (gene sets characterizing P1 and P2dominant samples) were externally validated in TCGA-COAD/READ (n = 624) and GSE39582 (n = 536), 1,160 cases in total, and therapeutic implications were tested in public cell line panels (GDSC2, DepMap) and in 65 lines of an independent patient derived CRC organoid biobank. Results: Deep learning morphological classification reached 98.5% test accuracy (64/65), and morphology corresponded to P1/P2 isoform usage: Type1 (compact glandular) morphology predominated in P1 dominant samples (median Type1 fraction 0.826 versus 0.444) and non Type1 (cystic mucinous) morphology in P2 dominant samples (AUC 0.79). Both signatures differed across the four consensus molecular subtypes , and the P2 signature was higher in mismatch repair deficient (microsatellite-unstable) tumors. Promoter usage quantified directly (P2_index) was higher in TP53 wild type tumors , consistent with P2 being p53-inducible. TP53 wild type cell lines were more sensitive to the MDM2 inhibitor Nutlin 3a and were more dependent on MDM2 in the DepMap CRISPR screen ; among 198 GDSC2 drugs, Nutlin 3a correlated most strongly with the P2 score. In the independent biobank, TP53 wild type lines (17) were more sensitive to nutlin-3 than mutant lines (48) (median log(IC50) 1.386 versus 4.283). Conclusions: MDM2 promoter choice (P1/P2) co-varies with the lineage identity of cancer cells and with the secretory, mucin rich character of the tumor tissue, consistent with a molecular-switch role alongside p53 suppression. The MDM2 P1/P2 ratio, measurable by RT qPCR or targeted NGS, is a candidate molecular-classification and therapeutic-stratification biomarker corresponding robustly to CMS, MSI, and TP53 mutation status.

bioinformatics↗