Post hoc attribution methods can provide insights into the learned patterns from deep neural networks (DNNs) trained on high-throughput functional genomics data. However, in practice, their resultant attribution maps can be challenging to interpret due to spurious importance scores for seemingly arbitrary nucleotides. Here, we identify a previously overlooked attribution noise source that arises from how DNNs handle one-hot encoded DNA. We demonstrate this noise is pervasive across various genomic DNNs and introduce a statistical correction that effectively reduces it, leading to more reliable attribution maps. Our approach represents a promising step towards gaining meaningful insights from DNNs in regulatory genomics.

Understanding coding mutations is important for many applications in biology and medicine but the vast mutation space makes comprehensive experimental characterisation impossible. Current predictors are often computationally intensive and difficult to scale, including recent deep learning models. We introduce Sequence UNET, a highly scalable deep learning architecture that classifies and predicts variant frequency from sequence alone using multi-scale representations from a fully convolutional compression/expansion architecture. It achieves comparable pathogenicity prediction to recent methods. We demonstrate scalability by analysing 8.3B variants in 904,134 proteins detected through large-scale proteomics. Sequence UNET runs on modest hardware with a simple Python package.

BACKGROUND: Genetic variation in regulatory sequences that alter transcription factor (TF) binding is a major cause of phenotypic diversity. Brassinosteroid is a growth hormone that has major effects on plant phenotypes. Genetic variation in brassinosteroid-responsive cis-elements likely contributes to trait variation. Pinpointing such regulatory variations and quantitative genomic analysis of the variation in TF-target binding, however, remains challenging. How variation in transcriptional targets of signaling pathways such as the brassinosteroid pathway contributes to phenotypic variation is an important question to be investigated with innovative approaches. RESULTS: Here, we use a hybrid allele-specific chromatin binding sequencing (HASCh-seq) approach and identify variations in target binding of the brassinosteroid-responsive TF ZmBZR1 in maize. HASCh-seq in the B73xMo17 F1s identifies thousands of target genes of ZmBZR1. Allele-specific ZmBZR1 binding (ASB) has been observed for 18.3% of target genes and is enriched in promoter and enhancer regions. About a quarter of the ASB sites correlate with sequence variation in BZR1-binding motifs and another quarter correlate with haplotype-specific DNA methylation, suggesting that both genetic and epigenetic variations contribute to the high level of variation in ZmBZR1 occupancy. Comparison with GWAS data shows linkage of hundreds of ASB loci to important yield and disease-related traits. CONCLUSION: Our study provides a robust method for analyzing genome-wide variations of TF occupancy and identifies genetic and epigenetic variations of the brassinosteroid response transcription network in maize.

Deep neural networks (DNNs) hold promise for functional genomics prediction, but their generalization capability may be limited by the amount of available data. To address this, we propose EvoAug, a suite of evolution-inspired augmentations that enhance the training of genomic DNNs by increasing genetic variation. Random transformation of DNA sequences can potentially alter their function in unknown ways, so we employ a fine-tuning procedure using the original non-transformed data to preserve functional integrity. Our results demonstrate that EvoAug substantially improves the generalization and interpretability of established DNNs across prominent regulatory genomics prediction tasks, offering a robust solution for genomic DNNs.

BACKGROUND: Plants memorize previous pathogen attacks and are "primed" to produce a faster and stronger defense response, which is critical for defense against pathogens. In plants, cytosines in transposons and gene bodies are reported to be frequently methylated. Demethylation of transposons can affect disease resistance by regulating the transcription of nearby genes during defense response, but the role of gene body methylation (GBM) in defense responses remains unclear. RESULTS: Here, we find that loss of the chromatin remodeler decrease in DNA methylation 1 (ddm1) synergistically enhances resistance to a biotrophic pathogen under mild chemical priming. DDM1 mediates gene body methylation at a subset of stress-responsive genes with distinct chromatin properties from conventional gene body methylated genes. Decreased gene body methylation in loss of ddm1 mutant is associated with hyperactivation of these gene body methylated genes. Knockout of glyoxysomal protein kinase 1 (gpk1), a hypomethylated gene in ddm1 loss-of-function mutant, impairs priming of defense response to pathogen infection in Arabidopsis. We also find that DDM1-mediated gene body methylation is prone to epigenetic variation among natural Arabidopsis populations, and GPK1 expression is hyperactivated in natural variants with demethylated GPK1. CONCLUSIONS: Based on our collective results, we propose that DDM1-mediated GBM provides a possible regulatory axis for plants to modulate the inducibility of the immune response.

We present a novel data structure for searching sequences in large databases: the Hierarchical Interleaved Bloom Filter (HIBF). It is extremely fast and space efficient, yet so general that it could serve as the underlying engine for many applications. We show that the HIBF is superior in build time, index size, and search time while achieving a comparable or better accuracy compared to other state-of-the-art tools. The HIBF builds an index up to 211 times faster, using up to 14 times less space, and can answer approximate membership queries faster by a factor of up to 129.

BACKGROUND: Genetic variation influences both chromatin accessibility, assessed in chromatin accessibility quantitative trait loci (caQTL) studies, and gene expression, assessed in expression QTL (eQTL) studies. Genetic variants can impact either nearby genes (cis-eQTLs) or distal genes (trans-eQTLs). Colocalization between caQTL and eQTL, or cis- and trans-eQTLs suggests that they share causal variants. However, pairwise colocalization between these molecular QTLs does not guarantee a causal relationship. Mediation analysis can be applied to assess the evidence supporting causality versus independence between molecular QTLs. Given that the function of QTLs can be cell-type-specific, we performed mediation analyses to find epigenetic and distal regulatory causal pathways for genes within two major cell types of the developing human cortex, progenitors and neurons. RESULTS: We find that the expression of 168 and 38 genes is mediated by chromatin accessibility in progenitors and neurons, respectively. We also find that the expression of 11 and 12 downstream genes is mediated by upstream genes in progenitors and neurons. Moreover, we discover that a genetic locus associated with inter-individual differences in brain structure shows evidence for mediation of SLC26A7 through chromatin accessibility, identifying molecular mechanisms of a common variant association to a brain trait. CONCLUSIONS: In this study, we identify cell-type-specific causal gene regulatory networks whereby the impacts of variants on gene expression were mediated by chromatin accessibility or distal gene expression. Identification of these causal paths will enable identifying and prioritizing actionable regulatory targets perturbing these key processes during neurodevelopment.

BACKGROUND: Transcriptional regulation is a key aspect of environmental stress responses. Heat stress induces transcriptional memory, i.e., sustained induction or enhanced re-induction of transcription, that allows plants to respond more efficiently to a recurrent HS. In light of more frequent temperature extremes due to climate change, improving heat tolerance in crop plants is an important breeding goal. However, not all heat stress-inducible genes show transcriptional memory, and it is unclear what distinguishes memory from non-memory genes. To address this issue and understand the genome and epigenome architecture of transcriptional memory after heat stress, we identify the global target genes of two key memory heat shock transcription factors, HSFA2 and HSFA3, using time course ChIP-seq. RESULTS: HSFA2 and HSFA3 show near identical binding patterns. In vitro and in vivo binding strength is highly correlated, indicating the importance of DNA sequence elements. In particular, genes with transcriptional memory are strongly enriched for a tripartite heat shock element, and are hallmarked by several features: low expression levels in the absence of heat stress, accessible chromatin environment, and heat stress-induced enrichment of H3K4 trimethylation. These results are confirmed by an orthogonal transcriptomic data set using both de novo clustering and an established definition of memory genes. CONCLUSIONS: Our findings provide an integrated view of HSF-dependent transcriptional memory and shed light on its sequence and chromatin determinants, enabling the prediction and engineering of genes with transcriptional memory behavior.

BACKGROUND: Therapy resistance in cancer is often driven by a subpopulation of cells that are temporarily arrested in a non-proliferative G0 state, which is difficult to capture and whose mutational drivers remain largely unknown. RESULTS: We develop methodology to robustly identify this state from transcriptomic signals and characterise its prevalence and genomic constraints in solid primary tumours. We show that G0 arrest preferentially emerges in the context of more stable, less mutated genomes which maintain TP53 integrity and lack the hallmarks of DNA damage repair deficiency, while presenting increased APOBEC mutagenesis. We employ machine learning to uncover novel genomic dependencies of this process and validate the role of the centrosomal gene CEP89 as a modulator of proliferation and G0 arrest capacity. Lastly, we demonstrate that G0 arrest underlies unfavourable responses to various therapies exploiting cell cycle, kinase signalling and epigenetic mechanisms in single-cell data. CONCLUSIONS: We propose a G0 arrest transcriptional signature that is linked with therapeutic resistance and can be used to further study and clinically track this state.

Single nucleotide variants (SNVs) contribute to human genomic diversity. Synonymous SNVs are previously considered to be "silent," but mounting evidence has revealed that these variants can cause RNA and protein changes and are implicated in over 85 human diseases and cancers. Recent improvements in computational platforms have led to the development of numerous machine-learning tools, which can be used to advance synonymous SNV research. In this review, we discuss tools that should be used to investigate synonymous variants. We provide supportive examples from seminal studies that demonstrate how these tools have driven new discoveries of functional synonymous SNVs.

BACKGROUND: Understanding the variation between well and poorly adapted cattle breeds to local environments and pathogens is essential for breeding cattle with improved climate and disease-resistant phenotypes. Although considerable progress has been made towards identifying genetic differences between breeds, variation at the epigenetic and chromatin levels remains poorly characterized. Here, we generate, sequence and analyse over 150 libraries at base-pair resolution to explore the dynamics of DNA methylation and chromatin accessibility of the bovine immune system across three distinct cattle lineages. RESULTS: We find extensive epigenetic divergence between the taurine and indicine cattle breeds across immune cell types, which is linked to the levels of local DNA sequence divergence between the two cattle sub-species. The unique cell type profiles enable the deconvolution of complex cellular mixtures using digital cytometry approaches. Finally, we show distinct sub-categories of CpG islands based on their chromatin and methylation profiles that discriminate between classes of distal and gene proximal islands linked to discrete transcriptional states. CONCLUSIONS: Our study provides a comprehensive resource of DNA methylation, chromatin accessibility and RNA expression profiles of three diverse cattle populations. The findings have important implications, from understanding how genetic editing across breeds, and consequently regulatory backgrounds, may have distinct impacts to designing effective cattle epigenome-wide association studies in non-European breeds.

Assay for Transposase-Accessible Chromatin with sequencing (ATAC-seq) reveals chromatin accessibility across the genome. Currently, no method specifically detects differential chromatin accessibility. Here, SeATAC uses a conditional variational autoencoder model to learn the latent representation of ATAC-seq V-plots and outperforms MACS2 and NucleoATAC on six separate tasks. Applying SeATAC to several pioneer factor-induced differentiation or reprogramming ATAC-seq datasets suggests that induction of these factors not only relaxes the closed chromatin but also decreases chromatin accessibility of 20% to 30% of their target sites. SeATAC is a novel tool to accurately reveal genomic regions with differential chromatin accessibility from ATAC-seq data.

BACKGROUND: Several models and algorithms have been proposed to build pangenomes from multiple input assemblies, but their impact on variant representation, and consequently downstream analyses, is largely unknown. RESULTS: We create multi-species super-pangenomes using pggb, cactus, and minigraph with the Bos taurus taurus reference sequence and eleven haplotype-resolved assemblies from taurine and indicine cattle, bison, yak, and gaur. We recover 221 k nonredundant structural variations (SVs) from the pangenomes, of which 135 k (61%) are common to all three. SVs derived from assembly-based calling show high agreement with the consensus calls from the pangenomes (96%), but validate only a small proportion of variations private to each graph. Pggb and cactus, which also incorporate base-level variation, have approximately 95% exact matches with assembly-derived small variant calls, which significantly improves the edit rate when realigning assemblies compared to minigraph. We use the three pangenomes to investigate 9566 variable number tandem repeats (VNTRs), finding 63% have identical predicted repeat counts in the three graphs, while minigraph can over or underestimate the count given its approximate coordinate system. We examine a highly variable VNTR locus and show that repeat unit copy number impacts the expression of proximal genes and non-coding RNA. CONCLUSIONS: Our findings indicate good consensus between the three pangenome methods but also show their individual strengths and weaknesses that need to be considered when analysing different types of variants from multiple input assemblies.

Genomics analyses use large reference sequence collections, like pangenomes or taxonomic databases. SPUMONI 2 is an efficient tool for sequence classification of both short and long reads. It performs multi-class classification using a novel sampled document array. By incorporating minimizers, SPUMONI 2's index is 65 times smaller than minimap2's for a mock community pangenome. SPUMONI 2 achieves a speed improvement of 3-fold compared to SPUMONI and 15-fold compared to minimap2. We show SPUMONI 2 achieves an advantageous mix of accuracy and efficiency in practical scenarios such as adaptive sampling, contamination detection and multi-class metagenomics classification.

We present RabbitTClust, a fast and memory-efficient genome clustering tool based on sketch-based distance estimation. Our approach enables efficient processing of large-scale datasets by combining dimensionality reduction techniques with streaming and parallelization on modern multi-core platforms. 113,674 complete bacterial genome sequences from RefSeq, 455 GB in FASTA format, can be clustered within less than 6 min and 1,009,738 GenBank assembled bacterial genomes, 4.0 TB in FASTA format, within only 34 min on a 128-core workstation. Our results further identify 1269 redundant genomes, with identical nucleotide content, in the RefSeq bacterial genomes database.

Spatial transcriptomics maps gene expression across tissues, posing the challenge of determining the spatial arrangement of different cell types. However, spatial transcriptomics spots contain multiple cells. Therefore, the observed signal comes from mixtures of cells of different types. Here, we propose an innovative probabilistic model, Celloscope, that utilizes established prior knowledge on marker genes for cell type deconvolution from spatial transcriptomics data. Celloscope outperforms other methods on simulated data, successfully indicates known brain structures and spatially distinguishes between inhibitory and excitatory neuron types based in mouse brain tissue, and dissects large heterogeneity of immune infiltrate composition in prostate gland tissue.

Computational methods represent the lifeblood of modern molecular biology. Benchmarking is important for all methods, but with a focus here on computational methods, benchmarking is critical to dissect important steps of analysis pipelines, formally assess performance across common situations as well as edge cases, and ultimately guide users on what tools to use. Benchmarking can also be important for community building and advancing methods in a principled way. We conducted a meta-analysis of recent single-cell benchmarks to summarize the scope, extensibility, and neutrality, as well as technical features and whether best practices in open data and reproducible research were followed. The results highlight that while benchmarks often make code available and are in principle reproducible, they remain difficult to extend, for example, as new methods and new ways to assess methods emerge. In addition, embracing containerization and workflow systems would enhance reusability of intermediate benchmarking results, thus also driving wider adoption.

Predicting the impact of coding and noncoding variants on splicing is challenging, particularly in non-canonical splice sites, leading to missed diagnoses in patients. Existing splice prediction tools are complementary but knowing which to use for each splicing context remains difficult. Here, we describe Introme, which uses machine learning to integrate predictions from several splice detection tools, additional splicing rules, and gene architecture features to comprehensively evaluate the likelihood of a variant impacting splicing. Through extensive benchmarking across 21,000 splice-altering variants, Introme outperformed all tools (auPRC: 0.98) for the detection of clinically significant splice variants. Introme is available at https://github.com/CCICB/introme .

BACKGROUND: The variation in the rate at which humans age may be rooted in early events acting through the genomic regions that are influenced by such events and subsequently are related to health phenotypes in later life. The parent-of-origin-effect (POE)-regulated methylome includes regions enriched for genetically controlled imprinting effects (the typical type of POE) and regions influenced by environmental effects associated with parents (the atypical POE). This part of the methylome is heavily influenced by early events, making it a potential route connecting early exposures, the epigenome, and aging. We aim to test the association of POE-CpGs with early and later exposures and subsequently with health-related phenotypes and adult aging. RESULTS: We perform a phenome-wide association analysis for the POE-influenced methylome using GS:SFHS (Ndiscovery = 5087, Nreplication = 4450). We identify and replicate 92 POE-CpG-phenotype associations. Most of the associations are contributed by the POE-CpGs belonging to the atypical class where the most strongly enriched associations are with aging (DNAmTL acceleration), intelligence, and parental (maternal) smoking exposure phenotypes. A proportion of the atypical POE-CpGs form co-methylation networks (modules) which are associated with these phenotypes, with one of the aging-associated modules displaying increased within-module methylation connectivity with age. The atypical POE-CpGs also display high levels of methylation heterogeneity, fast information loss with age, and a strong correlation with CpGs contained within epigenetic clocks. CONCLUSIONS: These results identify the association between the atypical POE-influenced methylome and aging and provide new evidence for the "early development of origin" hypothesis for aging in humans.

Enhancers are genomic DNA elements controlling spatiotemporal gene expression. Their flexible organization and functional redundancies make deciphering their sequence-function relationships challenging. This article provides an overview of the current understanding of enhancer organization and evolution, with an emphasis on factors that influence these relationships. Technological advancements, particularly in machine learning and synthetic biology, are discussed in light of how they provide new ways to understand this complexity. Exciting opportunities lie ahead as we continue to unravel the intricacies of enhancer function.

The Vertebrate Gene Nomenclature Committee (VGNC) was established in 2016 as a sister project to the HUGO Gene Nomenclature Committee, to approve gene nomenclature in vertebrate species without an existing dedicated nomenclature committee. The VGNC aims to harmonize gene nomenclature across selected vertebrate species in line with human gene nomenclature, with orthologs assigned the same nomenclature where possible. This article presents an overview of the VGNC project and discussion of key findings resulting from this work to date. VGNC-approved nomenclature is accessible at https://vertebrate.genenames.org and is additionally displayed by the NCBI, Ensembl, and UniProt databases.

BACKGROUND: Protein annotation is a major goal in molecular biology, yet experimentally determined knowledge is typically limited to a few model organisms. In non-model species, the sequence-based prediction of gene orthology can be used to infer protein identity; however, this approach loses predictive power at longer evolutionary distances. Here we propose a workflow for protein annotation using structural similarity, exploiting the fact that similar protein structures often reflect homology and are more conserved than protein sequences. RESULTS: We propose a workflow of openly available tools for the functional annotation of proteins via structural similarity (MorF: MorphologFinder) and use it to annotate the complete proteome of a sponge. Sponges are highly relevant for inferring the early history of animals, yet their proteomes remain sparsely annotated. MorF accurately predicts the functions of proteins with known homology in [Formula: see text] cases and annotates an additional [Formula: see text] of the proteome beyond standard sequence-based methods. We uncover new functions for sponge cell types, including extensive FGF, TGF, and Ephrin signaling in sponge epithelia, and redox metabolism and control in myopeptidocytes. Notably, we also annotate genes specific to the enigmatic sponge mesocytes, proposing they function to digest cell walls. CONCLUSIONS: Our work demonstrates that structural similarity is a powerful approach that complements and extends sequence similarity searches to identify homologous proteins over long evolutionary distances. We anticipate this will be a powerful approach that boosts discovery in numerous -omics datasets, especially for non-model organisms.

BACKGROUND: Plant architecture associated with increased grain yield and adaptation to the local environments is selected during wheat (Triticum aestivum) breeding. The internode length of individual stems and tiller length of individual plants are important for the determination of plant architecture. However, few studies have explored the genetic basis of these traits. RESULTS: Here, we conduct a genome-wide association study (GWAS) to dissect the genetic basis of geographical differentiation of these traits in 306 worldwide wheat accessions including both landraces and traditional varieties. We determine the changes of haplotypes for the associated genomic regions in frequency in 831 wheat accessions that are either introduced from other countries or developed in China from last two decades. We identify 83 loci that are associated with one trait, while the remaining 247 loci are pleiotropic. We also find 163 associated loci are under strong selective sweep. GWAS results demonstrate independent regulation of internode length of individual stems and consistent regulation of tiller length of individual plants. This makes it possible to obtain ideal haplotype combinations of the length of four internodes. We also find that the geographical distribution of the haplotypes explains the observed differences in internode length among the worldwide wheat accessions. CONCLUSION: This study provides insights into the genetic basis of plant architecture. It will facilitate gene functional analysis and molecular design of plant architecture for breeding.

BACKGROUND: Verticillium wilt is one of the most devasting diseases for many plants, leading to global economic loss. Cotton is known to be vulnerable to its fungal pathogen, Verticillium dahliae, yet the related genetic mechanism remains unknown. RESULTS: By genome-wide association studies of 419 accessions of the upland cotton, Gossypium hirsutum, we identify ten loci that are associated with resistance against Verticillium wilt. Among these loci, SHZDI1/SHZDP2/AYDP1 from chromosome A10 is located on a fragment introgressed from Gossypium arboreum. We characterize a large cluster of Toll/interleukin 1 (TIR) nucleotide-binding leucine-rich repeat receptors in this fragment. We then identify a dual-TIR domain gene from this cluster, GhRVD1, which triggers an effector-independent cell death and is induced by Verticillium dahliae. We confirm that GhRVD1 is one of the causal gene for SHZDI1. Allelic variation in the TIR domain attenuates GhRVD1-mediated resistance against Verticillium dahliae. Homodimerization between TIR1-TIR2 mediates rapid immune response, while disruption of its αD- and αE-helices interface eliminates the autoactivity and self-association of TIR1-TIR2. We further demonstrate that GhTIRP1 inhibits the autoactivity and self-association of TIR1-TIR2 by competing for binding to them, thereby preventing the resistance to Verticillium dahliae. CONCLUSIONS: We propose the first working model for TIRP1 involved self-association and autoactivity of dual-TIR domain proteins that confer compromised pathogen resistance of dual-TIR domain proteins in plants. The findings reveal a novel mechanism on Verticillium dahliae resistance and provide genetic basis for breeding in future.