This content will become publicly available on March 27, 2027

Title: Integrating protein and DNA embeddings for improving genome-wide transcription factor binding site prediction
Abstract Transcription factors (TFs) regulate gene expression by binding to specific DNA sites on genome, making accurate TF binding site prediction critical for understanding gene regulation and downstream phenotypes. Almost all current deep learning methods use only DNA-related information to predict TF binding sites, ignoring the fact that different TF protein sequences and structures recognize distinct DNA patterns. Not leveraging TF information not only limits prediction accuracy but also makes the methods not generalizable to predicting binding sites of new TFs that do not exist in the training data. Here, we present TransBind, a protein-aware deep learning architecture that integrates DNA sequence information with protein embeddings containing both sequence and structural information derived from a protein language model pretrained on DNA-binding proteins, to improve TF binding site prediction. Through the cross-attention, a TF embedding selectively attends to genomic regions according to its unique binding properties. Evaluated on the data of 690 ChIP-seq experiments spanning 161 TFs across 91 human cell types, TransBind achieves an AUROC of 0.9508 and AUPR of 0.3741—representing a $$\ge$$11.8% relative AUPR improvement over state-of-the-art methods including TBiNet, EPBDXDNABERT-2, DanQ, and DeepSEA. The model outperformed existing methods in $$\ge$$98% of TF–cell type combinations. It also recovered 160 known TF binding motifs in the JASPAR database, providing the biological interpretability of the model. Moreover, the approach enables label-zero-shot prediction for unseen TFs, demonstrating its potential of generalizing to new, poorly characterized TFs. The source code of TransBind is available at https://github.com/jianlin-cheng/TransBind. The version used in this work is archived at https://doi.org/10.5281/zenodo.19462292.  more » « less
Award ID(s):
2343612
PAR ID:
10691273
Author(s) / Creator(s):
;
Publisher / Repository:
Oxford
Date Published:
Journal Name:
NAR Genomics and Bioinformatics
Volume:
8
Issue:
2
ISSN:
2631-9268
Subject(s) / Keyword(s):
Sponsoring Org:
National Science Foundation
More Like this
  1. Salz, H (Ed.)
    Abstract During gene regulation, DNA accessibility is thought to limit the availability of transcription factor (TF) binding sites, while TFs can increase DNA accessibility to recruit additional factors that upregulate gene expression. Given this interplay, the causative regulatory events in the modulation of gene expression remain unknown for the vast majority of genes. We utilized deeply sequenced ATAC-Seq data and site-specific knock-in reporter genes to investigate the relationship between the binding-site resolution dynamics of DNA accessibility and the expression dynamics of the enhancers of Cebpa during macrophage-neutrophil differentiation. While the enhancers upregulate reporter expression during the earliest stages of differentiation, there is little corresponding increase in their total accessibility. Conversely, total accessibility peaks during the last stages of differentiation without any increase in enhancer activity. The accessibility of positions neighboring C/EBP-family TF binding sites, which indicates TF occupancy, does increase significantly during early differentiation, showing that the early upregulation of enhancer activity is driven by TF binding. These results imply that a generalized increase in DNA accessibility is not sufficient, and binding by enhancer-specific TFs is necessary, for the upregulation of gene expression. Additionally, high-coverage ATAC-Seq combined with time-series expression data can infer the sequence of regulatory events at binding-site resolution. 
    more » « less
  2. Abstract xBind is an interactive, freely accessible, and fully configurable webserver for large language model (LLM)-enabled cross-molecular protein binding-site prediction. xBind leverages LLM embeddings from the ESM-2 model together with sequence- and structure-derived features to predict protein–protein, protein–DNA, and protein–RNA binding sites using symmetry-aware deep graph neural networks. The input to xBind is either a single-chain protein sequence in FASTA format or a monomer protein structure in PDB or mmCIF format and it outputs predicted residue-level binding sites of the input protein with its pre-selected interaction partner. The customizable xBind web interface provides: (i) choice of interaction partners including protein–protein, protein–DNA, and protein–RNA; (ii) on-the-fly AlphaFold-based protein structure prediction for sequence-only inputs; (iii) on-demand selection of the likelihood threshold for calibrating structure-aware binding site annotations; (iv) interactive and interpretable web-based results, including sequence and structural visualizations and plots of residue-level binding likelihoods with user-adjustable threshold calibration; and (v) extensive help information for usage and results interpretation through a web-based tutorial and guide. xBind is freely available at https://fusion.cs.vt.edu/xBind. 
    more » « less
  3. Transcription factors (TFs) are essential regulators of gene expression, and variations in their target DNA sequences due to altering TF-DNA binding affinity and specificity lead to diseases ranging from developmental disorders to cancer. Computational methods that integrate physics-based models with machine learning (ML) hold promise to accurately predict protein−DNA binding affinities while ensuring interpretability and generalizability. Here, we present an approach combining all-atom molecular dynamics (MD) simulations and Molecular Mechanics-Generalized Born Surface Area (MMGBSA) energy calculations with ML model constructions (neural networks, random forests, and support vector machines) to predict DNA binding affinities and specificities for the dimeric TF Myc/Max. Using high-quality experimental data from genomic- context protein-binding microarrays (gcPBM), we constructed a balanced data set of 168 DNA sequences reflecting physiologically relevant genomic environments. Multiple independent simulations were conducted per sequence for each TF-DNA complex to capture structural dynamic and interaction properties, with physically essential energetic descriptors extracted, including van der Waals, electrostatic, solvation, hydrogen bonding, and additional energy corrections. Our models achieved a Pearson correlation of∼0.73 and a mean absolute error of 0.4, substantially improving upon conventional MMGBSA prediction. Feature importance analyses highlighted TF-DNA interfacial complementarity and hydrophobic interactions as primary determinants of binding affinity and specificity, though TF-DNA interfacial hydrogen bonding contributions remain to be better characterized physically for sequence dependency. This physics-informed ML framework thus aims at both predictive accuracy and mechanistic interpretability, paving the way toward universal scalable prediction of interpretable protein−DNA interactions. 
    more » « less
  4. The intrinsic DNA sequence preferences and cell type–specific cooperative partners of transcription factors (TFs) are typically highly conserved. Hence, despite the rapid evolutionary turnover of individual TF binding sites, predictive sequence models of cell type–specific genomic occupancy of a TF in one species should generalize to closely matched cell types in a related species. To assess the viability of cross-species TF binding prediction, we train neural networks to discriminate ChIP-seq peak locations from genomic background and evaluate their performance within and across species. Cross-species predictive performance is consistently worse than within-species performance, which we show is caused in part by species-specific repeats. To account for this domain shift, we use an augmented network architecture to automatically discourage learning of training species–specific sequence features. This domain adaptation approach corrects for prediction errors on species-specific repeats and improves overall cross-species model performance. Our results show that cross-species TF binding prediction is feasible when models account for domain shifts driven by species-specific repeats. 
    more » « less
  5. Abstract Transcription factors (TF) are proteins that bind DNA in a sequence-specific manner to regulate gene transcription. Despite their unique intrinsic sequence preferences,in vivogenomic occupancy profiles of TFs differ across cellular contexts. Hence, deciphering the sequence determinants of TF binding, both intrinsic and context-specific, is essential to understand gene regulation and the impact of regulatory, non-coding genetic variation. Biophysical models trained onin vitroTF binding assays can estimate intrinsic affinity landscapes and predict occupancy based on TF concentration and affinity. However, these models cannot adequately explain context-specific,in vivobinding profiles. Conversely, deep learning models, trained onin vivoTF binding assays, effectively predict and explain genomic occupancy profiles as a function of complex regulatory sequence syntax, albeit without a clear biophysical interpretation. To reconcile these complementary models ofin vitroandin vivoTF binding, we developed Affinity Distillation (AD), a method that extracts thermodynamic affinitiesde-novofrom deep learning models of TF chromatin immunoprecipitation (ChIP) experiments by marginalizing away the influence of genomic sequence context. Applied to neural networks modeling diverse classes of yeast and mammalian TFs, AD predicts energetic impacts of sequence variation within and surrounding motifs on TF binding as measured by diversein vitroassays with superior dynamic range and accuracy compared to motif-based methods. Furthermore, AD can accurately discern affinities of TF paralogs. Our results highlight thermodynamic affinity as a key determinant ofin vivobinding, suggest that deep learning models ofin vivobinding implicitly learn high-resolution affinity landscapes, and show that these affinities can be successfully distilled using AD. This new biophysical interpretation of deep learning models enables high-throughputin silicoexperiments to explore the influence of sequence context and variation on both intrinsic affinity andin vivooccupancy. 
    more » « less