Aug 2026· PLoS Computational Biology· Vol 22, pp. e1014626· 0 citations· 55 references
Medicine
TL;DR
EvoSNR-Prom is built on the Evo foundation model and formulates promoter identification as a token-level sequence labeling problem, analogous to named entity recognition in natural language processing, and a lexicon-enhanced embedding strategy is introduced that incorporates biologically meaningful DNA lexicons, enriching contextual representations and improving the model’s ability to capture complex sequence motifs.
Abstract
The precise identification of promoters is crucial for understanding gene regulation. Deep learning methods have achieved considerable success in promoter prediction, yet most operate at the sequence level with coarse-grained labels. This means they label an entire DNA segment as either a “promoter” or “non-promoter,” which results in a lack of the nucleotide-level resolution in prediction. In this study, we propose EvoSNR-Prom, a model designed for promoter prediction at single-nucleotide resolution. EvoSNR-Prom is built on the Evo foundation model and formulates promoter identification as a token-level sequence labeling problem, analogous to named entity recognition in natural language processing. To address the limited contextual information available in single-nucleotide tokenization, we introduce a lexicon-enhanced embedding strategy that incorporates biologically meaningful DNA lexicons, enriching contextual representations and improving the model’s ability to capture complex sequence motifs. Furthermore, to enhance predictive performance on small size datasets, we integrate a label-aware transfer learning framework to leverage knowledge from well-annotated source species to a target organism. The results across various prokaryotic datasets show that EvoSNR-Prom achieves excellent performance. This work provides a valuable computational framework for the high-precision analysis of gene regulatory elements, contributing to the advancement of promoter prediction at single-nucleotide resolution.
Identifying transcription factor binding sites (TFBSs) is fundamental to understanding complex gene regulatory mechanisms and the functions of non-coding regions. Although existing methods have achieved substantial strides, capturing both local structural features and long-range spatial dependencies within DNA sequences remains a major challenge for improving prediction accuracy. In this study, we propose DNCLA, a deep learning model that synergizes multisize convolutional fusion, Bidirectional Long ShortTerm Memory (Bi-LSTM) networks, and a multi-head self-attention
mechanism. At the feature extraction level, DNCLA breaks through the limitations of traditional single-sequence encoding by fusing Nucleotide Chemical Properties (NCP) with Dinucleotide Physicochemical Properties (DPCP). NCP provides a refined characterization of chemical differences between bases based on ring structures, hydrogen bond sites, and functional group properties, while DPCP introduces parameters such as local structural stability and geometric flexibility of the DNA. Subsequently, the model extracts spatial evolution from these high-dimensional features through a multi-size convolutional module; captures long-range spatial dependencies using Bi-LSTM layers; and employs a multi-head self-attention mechanism to achieve adaptive weight distribution of global features, thereby enhancing the perception of key regulatory motifs. Results from training and testing the proposed model on 165 ChIPseq datasets demonstrate that DNCLA possesses robust generalization capabilities and high predictive performance in TFBSs identification. This suggests that the incorporation of physicochemical features better elucidates the essence of interactions between transcription factors and DNA.
Jingjue Wei, Jie Feng· Match-communications in Math...· 0 citations
GB.GeneUnet, an 837M-parameter transformer-based U-Net pretrained on 6 trillion tokens from multi-species genomes in OpenGenome2 is introduced, extending genomic context to 1 Mb with up to 100× inference speedup over GeneMoE, a preliminary MoE transformer baseline of similar model size pretrained on the same data.
Ning Sun, William de Vazelhes, Pan Li et al.· bioRxiv· 0 citations
Gene expression is governed by regulatory DNA and their associated trans factors acting in specific cell types, yet the sequences underlying this control remain poorly mapped in plants. Genome-pretrained DNA language models provide a route to interrogate regulatory sequence directly, but their attributions have largely been interpreted using bulk or whole-tissue data, and standard attribution pipelines can preferentially highlight sequences downstream of the transcription start (TSS) site rather than promoter-associated signals. Here, we train a celltype-resolved sequence-to-expression model from a single-cell soybean (Glycine max) atlas by coupling a soybean-adapted Genomic Pre-trained Network (GPN) to a shared sequence encoder with 66 cell-type-specific output heads. Across 38,339 protein-coding genes, the model achieves a mean per-cell-type, across-gene Pearson correlation of 0.683 and, recast as a highversus-low expression classification, reaches an area under the ROC curve of 0.92 to 0.97 across tissues, at or above dedicated plant sequence models. We then introduce ContextAware Significance of Cross-gene Attribution for Discovering Elements (CASCADE), a positionspecific statistical framework for identifying model-derived candidate regulatory elements from in silico saturation mutagenesis. Relative to the pooled null used by TF-MoDISco, CASCADE shifts motif recovery from downstream of the transcription start site toward promoter sequence, with 77% of CASCADE-exclusive motifs, compared with 12% of TF-MoDISco-exclusive motifs, falling within the promoter. Applied across the atlas, CASCADE identifies approximately 1.39 million candidate elements spanning broadly active, tissue-restricted and cell-type-restricted classes. Together, these analyses establish a position-aware approach for extracting promoterassociated regulatory hypotheses from sequence models and generate a cell-type-resolved map of candidate cis-regulatory elements.
Ali Farghadan, Robert J. Schmitz, Scott A. Jackson et al.· bioRxiv· 0 citations
MOTIVATION
Genomic sequence-to-activity models can decipher gene regulatory mechanisms and predict the functional impact of regulatory variants. However, current models struggle to integrate information from sequences outside promoters, especially information from cell type specific regulatory elements.
RESULTS
Here, we propose incorporating base-pair resolution evolutionary conservation data into genomic sequence-to-expression predictors. We explore two training strategies-training from scratch or fine-tuning an existing sequence-only model with additional conservation input. We find that in both cases, base-pair resolution conservation data improves cell type specific sequence-to-expression prediction, with training from scratch yielding the greatest benefit. The improvement in cell type specific expression prediction can be attributed in part to the fact that models trained on sequence and conservation data learn to better recognize cell type specific regulatory elements than models trained on sequence alone.
AVAILABILITY
Code is available at https://github.com/ni-lab/basenji-phyloP.
SUPPLEMENTARY INFORMATION
Supplementary data are available at Bioinformatics online.
Pooja Kathail, Forest L. Yang, Gabriel B. Loeb et al.· Bioinformatics· 0 citations
PLM-ArgMe is presented that is based on a symmetry-sensitive Transformer framework using context-aware ESM-2 residue embeddings, which is mapped through a novel Bio-Symmetric Mirrored Sinusoidal Encoding strategy to address the biological symmetry hypothesis of arginine methylation.
Nitika Bhatt, Kartik Joshi, R. Rout et al.· Biochemical and Biophysical...· 0 citations
Abstract Recent advances in large language models have extended to genomic applications, yet model robustness relative to context is unclear. Here, we demonstrate two intrinsic biases (input sequence length and nucleotide position) affecting SegmentNT results, a model included with the Nucleotide Transformer that provides nucleotide-level predictions of biological features. We demonstrate that nucleotide position within the input sequence (beginning, middle, or end) alters the nature of SegmentNT’s raw prediction probabilities, which can be standardized to improve prediction consistency. While longer input sequence length improves model performance, diminishing returns suggest a surprisingly small input length of ∼3072 nucleotides might be sufficient for many applications. We further identify a 24-nucleotide periodic oscillation in SegmentNT’s prediction probabilities, revealing an intrinsic bias potentially linked to the model’s training tokenization (6-mers) and architecture. We identify potential approaches to account for these biases and provide generalizable insights for utilizing nucleotide-resolution functional prediction models.
M. Ebbert, Anna Ho, Madeline L. Page et al.· Nucleic Acids Research· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
