Self-supervised DNA language models (DNALMs) are typically trained at massive scale on whole genomes and long contexts. However, regulatory sequence features are sparse, heterogeneous, and dominated by poorly conserved flexible syntax of short motifs, which can be difficult to learn from genome-wide self-supervision. As a result, annotation agnostic, long-context DNALMs struggle to learn regulatory syntax and can underperform simpler baseline models on key regulatory tasks. We therefore introduce ARSENAL, a short-context masked DNA language model trained on a functionally enriched regulatory corpus and augmented with a novel regularizer than that encourages motif discovery. ARSENAL improves recovery of diverse transcription factor motifs de novo and prediction of regulatory variant effects in the zero-shot setting compared to other DNALMs. Incorporating ARSENAL embeddings also improves supervised chromatin accessibility prediction over strong ab-initio baselines across multiple cell types and yields improved regulatory variant scoring. Finally, ARSENAL serves as a practical generative prior, enabling targeted regulatory sequence design under downstream functional constraints. All code can be found at https://github.com/kundajelab/regulatory_lm, and models and data can be found at https://sageb.io/4ZpEnk.

Cell types can be organized into related families, but the regulatory mechanisms that define and maintain these families across deep evolutionary time remain unknown. Here, combining single-nucleus multi-omic sequencing with deep learning to analyse the accessible genomes of two groups of vastly divergent animals including flatworms and vertebrates, we find that hundreds of accessibility-dictating sequence motifs partition into distinct yet conserved sets, or 'vocabularies', each associated with a specific cell type family. However, combinatorial relationships among these motifs preferred by individual cell types are largely species specific. Deep-learning models trained on one species accurately predict family-level chromatin accessibility in distantly related species, albeit frequently rely on different motifs from shared vocabularies to reach convergent predictions. By contrast, models trained on individual cell types within a family lose cross-species predictive power, indicating that the regulatory syntax governing cell type-level identity evolves rapidly. We propose a 'collective maintenance' model in which motif vocabularies defining cell type families are evolutionarily stable, while recombination of these motifs generates cell type-specific regulatory programmes. This suggests that family identity is maintained collectively by large, conserved pools of regulatory factors, analogous to the logic of developmental homology, where character identity persists through network-level conservation despite extensive rewiring.

JASPAR (https://jaspar.elixir.no/) is an open-access database that has provided high-quality, manually curated, and non-redundant DNA binding profiles for transcription factors (TFs) as position frequency matrices (PFMs) for over 20 years. We expanded the CORE (306 new profiles, 12% increase) and UNVALIDATED (433, 60% increase) collections with new PFMs and updated 13 existing profiles. We updated the TF binding site predictions and genome tracks for eight species. TF binding profile clusters and familial TF binding sites were updated accordingly. We integrate the inMOTIFin software to easily simulate regulatory sequences using JASPAR PFMs. To enrich TFs' annotations, we provide scientific literature-based human TF target information. Notably, this release features a deep learning (DL) collection, providing a paradigm shift in modeling and characterizing TF-DNA interactions with 1259 BPNet models trained on Homo sapiens ENCODE chromatin immunoprecipitation followed by sequencing (ChIP-seq) datasets from 240 TFs and interpreted to reveal predictive motif patterns for the models. The motifs associated with the same TF were clustered to provide a summary of the binding properties, resulting in 240 primary and 113 alternative motif patterns in the DL collection. The JASPAR 2026 collections lay a foundation for future endeavors in genomic research, serving the scientific community in uncovering the mechanisms of gene regulation.

Recent advances in self-supervised models for natural language, vision, and protein sequences have catalyzed the development of genomic DNA language models (DNALMs). These models aim to learn generalizable representations of diverse DNA elements, potentially enabling various downstream genomic prediction, interpretation and design tasks. However, existing benchmarks do not adequately assess the capabilities of DNALMs on an important class of non-coding DNA elements critical for regulating gene activity. Here, we introduce DART-Eval, a suite of representative benchmarks focused on regulatory DNA to evaluate performance of DNALMs across zero-shot, probed, and fine-tuned settings against contemporary ab initio models as baselines. DART-Eval addresses biologically relevant tasks including sequence motif discovery, cell-type specific regulatory activity prediction, and counterfactual prediction of regulatory genetic variants. Our systematic evaluations reveal that current annotation-agnostic DNALMs exhibit inconsistent performance and do not offer compelling gains over alternative baseline models for most tasks, despite requiring significantly more computational resources. We discuss potentially promising modeling, data curation, and evaluation strategies for the next generation of DNALMs. Our benchmark datasets and evaluation framework are available at https://github.com/kundajelab/DART-Eval.

Cell types evolve into a hierarchy with related types grouped into families. How cell type diversification is constrained by the stable separation between families over vast evolutionary times remains unknown. Here, integrating single-nucleus multiomic sequencing and deep learning, we show that hundreds of sequence features (motifs) divide into distinct sets associated with accessible genomes of specific cell type families. This division is conserved across highly divergent, early-branching animals including flatworms and cnidarians. While specific interactions between motifs delineate cell type relationships within families, surprisingly, these interactions are not conserved between species. Consistently, while deep learning models trained on one species can predict accessibility of other species' sequences, their predictions frequently rely on distinct, but synonymous, motif combinations. We propose that long-term stability of cell type families is maintained through genome access specified by conserved motif sets, or 'vocabularies', whereas cell types diversify through flexible use of motifs within each set.

The genome is a sequence that completely encodes the DNA, RNA, and proteins that orchestrate the function of a whole organism. Advances in machine learning combined with massive datasets of whole genomes could enable a biological foundation model that accelerates the mechanistic understanding and generative design of complex molecular interactions. We report Evo, a genomic foundation model that enables prediction and generation tasks from the molecular to genome scale. Using an architecture based on advances in deep signal processing, we scale Evo to 7 billion parameters with a context length of 131 kilobases (kb) at single-nucleotide, byte resolution. Trained on whole prokaryotic genomes, Evo can generalize across the three fundamental modalities of the central dogma of molecular biology to perform zero-shot function prediction that is competitive with, or outperforms, leading domain-specific language models. Evo also excels at multi-element generation tasks, which we demonstrate by generating synthetic CRISPR-Cas molecular complexes and entire transposable systems for the first time. Using information learned over whole genomes, Evo can also predict gene essentiality at nucleotide resolution and can generate coding-rich sequences up to 650 kb in length, orders of magnitude longer than previous methods. Advances in multi-modal and multi-scale learning with Evo provides a promising path toward improving our understanding and control of biology across multiple levels of complexity.

Genomic (DNA) sequences encode an enormous amount of information for gene regulation and protein synthesis. Similar to natural language models, researchers have proposed foundation models in genomics to learn generalizable features from unlabeled genome data that can then be fine-tuned for downstream tasks such as identifying regulatory elements. Due to the quadratic scaling of attention, previous Transformer-based genomic models have used 512 to 4k tokens as context (<0.001% of the human genome), significantly limiting the modeling of long-range interactions in DNA. In addition, these methods rely on tokenizers or fixed k-mers to aggregate meaningful DNA units, losing single nucleotide resolution where subtle genetic variations can completely alter protein function via single nucleotide polymorphisms (SNPs). Recently, Hyena, a large language model based on implicit convolutions was shown to match attention in quality while allowing longer context lengths and lower time complexity. Leveraging Hyena's new long-range capabilities, we present HyenaDNA, a genomic foundation model pretrained on the human reference genome with context lengths of up to 1 million tokens at the single nucleotide-level - an up to 500x increase over previous dense attention-based models. HyenaDNA scales sub-quadratically in sequence length (training up to 160x faster than Transformer), uses single nucleotide tokens, and has full global context at each layer. We explore what longer context enables - including the first use of in-context learning in genomics. On fine-tuned benchmarks from the Nucleotide Transformer, HyenaDNA reaches state-of-the-art (SotA) on 12 of 18 datasets using a model with orders of magnitude less parameters and pretraining data. On the GenomicBenchmarks, HyenaDNA surpasses SotA on 7 of 8 datasets on average by +10 accuracy points. Code at https://github.com/HazyResearch/hyena-dna.

Excel file of Supplementary Tables

Genetic variants associated with complex traits are primarily noncoding, and their effects on gene-regulatory activity remain largely uncharacterized. To address this, we profile epigenomic variation of histone mark H3K27ac across 387 brain, heart, muscle and lung samples from Genotype-Tissue Expression (GTEx). We annotate 282 k active regulatory elements (AREs) with tissue-specific activity patterns. We identify 2,436 sex-biased AREs and 5,397 genetically influenced AREs associated with 130 k genetic variants (haQTLs) across tissues. We integrate genetic and epigenomic variation to provide mechanistic insights for disease-associated loci from 55 genome-wide association studies (GWAS), by revealing candidate tissues of action, driver SNPs and impacted AREs. Lastly, we build ARE–gene linking scores based on genetics (gLink scores) and demonstrate their unique ability to prioritize SNP–ARE–gene circuits. Overall, our epigenomic datasets, computational integration and mechanistic predictions provide valuable resources and important insights for understanding the molecular basis of human diseases/traits such as schizophrenia.

Studies of optical properties of doped nanocrystals of tungsten trioxide can elucidate new information about the material. A novel molecule-enhanced photoluminescence (PL) of potassium-doped tungsten trioxide (K x WO) was explored in the presence of different gases to understand charge transfer between molecules and K x WO on the properties of the material. We performed Raman spectroscopy and PL experiments in the presence of gaseous acetone or ethanol mixed with other gases (N2 and O2). PL at 630 nm from K x WO was observed and further enhanced when the sample was continuously irradiated with a 532 nm CW laser in acetone. A mechanism of strong emission of the PL induced by the charge transfer between the acetone and the K x WO is proposed.

Genome-wide association studies (GWASs) of eye disorders have identified hundreds of genetic variants associated with ocular disease. However, the vast majority of these variants are noncoding, making it challenging to interpret their function. Here we present a joint single-cell atlas of gene expression and chromatin accessibility of the adult human retina with more than 50,000 cells, which we used to analyze single-nucleotide polymorphisms (SNPs) implicated by GWASs of age-related macular degeneration, glaucoma, diabetic retinopathy, myopia, and type 2 macular telangiectasia. We integrate this atlas with a HiChIP enhancer connectome, expression quantitative trait loci (eQTL) data, and base-resolution deep learning models to predict noncoding SNPs with causal roles in eye disease, assess SNP impact on transcription factor binding, and define their known and novel target genes. Our efforts nominate pathogenic SNP-target gene interactions for multiple vision disorders and provide a potentially powerful resource for interpreting noncoding variation in the eye.

The production of population-level trees using the genomic data of individuals is a fundamental task in the field of population genetics. Typically, these trees are produced using methods like hierarchical clustering, neighbor joining, or maximum likelihood. However, such methods are non-parametric: they require all data to be present at the time of tree formation, and the addition of new data points necessitates the regeneration of the entire tree, a potentially expensive process. They also do not easily integrate with larger workflows. In this study, we aim to address these problems by introducing parametric deep learning methods for tree formation from genotype data. Our models specifically create continuous representations of population trees in hyperbolic space, which has previously proven highly effective in embedding hierarchically structured data. We present two different architectures - a multi-layer perceptron (MLP) and a variational autoencoder (VAE) - and we analyze their performance using a variety of metrics along with comparisons to established tree-building methods. Both models tested produce embedding spaces that reflect human evolutionary history. In addition, we demonstrate the generalizability of these models by verifying that addition of new samples to an existing tree occurs in a semantically meaningful manner. Finally, we use Dasgupta’s Cost to compare the quality of trees generated by our models to those produced by established methods. Despite the fact that the benchmark methods are directly fit on the evaluation data, our models are able to outperform some of these and achieve highly comparable performance overall. Author summary Tree production is a vital task in population genetics, but current approaches fall prey to several common shortfalls. Most notably, they lack the ability to add new data points after tree generation, and they are often difficult to use in larger pipelines. By leveraging cutting-edge advances pairing deep learning with hyperbolic geometry, we develop multiple models designed to rectify these issues. Through experiments on a dataset of humans from globally widespread ancestries, we demonstrate the generalizability of our models to new data, and we also show strong empirical performance with respect to currently used methods. In addition, we show that the data representations produced by our models are semantically meaningful and reflect known facts about human evolutionary history. Finally, we discuss the additional benefits our models could provide, including improved visualization, greater privacy preservation, and improved integration with downstream machine learning tasks. In conclusion, we present models that are accurate, flexible, and generalizable, with the potential to facilitate a variety of further applications.

Abstract Effective data sharing is key to accelerating research to improve diagnostic precision, treatment efficacy, and long-term survival in pediatric cancer and other childhood catastrophic diseases. We present St. Jude Cloud (https://www.stjude.cloud), a cloud-based data-sharing ecosystem for accessing, analyzing, and visualizing genomic data from >10,000 pediatric patients with cancer and long-term survivors, and >800 pediatric sickle cell patients. Harmonized genomic data totaling 1.25 petabytes are freely available, including 12,104 whole genomes, 7,697 whole exomes, and 2,202 transcriptomes. The resource is expanding rapidly, with regular data uploads from St. Jude's prospective clinical genomics programs. Three interconnected apps within the ecosystem—Genomics Platform, Pediatric Cancer Knowledgebase, and Visualization Community—enable simultaneously performing advanced data analysis in the cloud and enhancing the Pediatric Cancer knowledgebase. We demonstrate the value of the ecosystem through use cases that classify 135 pediatric cancer subtypes by gene expression profiling and map mutational signatures across 35 pediatric cancer subtypes. Significance: To advance research and treatment of pediatric cancer, we developed St. Jude Cloud, a data-sharing ecosystem for accessing >1.2 petabytes of raw genomic data from >10,000 pediatric patients and survivors, innovative analysis workflows, integrative multiomics visualizations, and a knowledgebase of published data contributed by the global pediatric cancer community. This article is highlighted in the In This Issue feature, p. 995

Abstract Genomic studies of pediatric cancer have primarily focused on specific tumor types or high-risk disease. Here, we used a three-platform sequencing approach, including whole-genome sequencing (WGS), whole-exome sequencing (WES), and RNA sequencing (RNA-seq), to examine tumor and germline genomes from 309 prospectively identified children with newly diagnosed (85%) or relapsed/refractory (15%) cancers, unselected for tumor type. Eighty-six percent of patients harbored diagnostic (53%), prognostic (57%), therapeutically relevant (25%), and/or cancer-predisposing (18%) variants. Inclusion of WGS enabled detection of activating gene fusions and enhancer hijacks (36% and 8% of tumors, respectively), small intragenic deletions (15% of tumors), and mutational signatures revealing of pathogenic variant effects. Evaluation of paired tumor–normal data revealed relevance to tumor development for 55% of pathogenic germline variants. This study demonstrates the power of a three-platform approach that incorporates WGS to interrogate and interpret the full range of genomic variants across newly diagnosed as well as relapsed/refractory pediatric cancers. Significance: Pediatric cancers are driven by diverse genomic lesions, and sequencing has proven useful in evaluating high-risk and relapsed/refractory cases. We show that combined WGS, WES, and RNA-seq of tumor and paired normal tissues enables identification and characterization of genetic drivers across the full spectrum of pediatric cancers. This article is highlighted in the In This Issue feature, p. 2945

Introduction: Clinicians in academia face four major career challenges:

Abstract Accurate tissue of origin (TOO) prediction is crucial for effective clinical follow-up in early cancer detection from blood. In the second substudy of the Circulating Cell-free Genome Atlas (CCGA; NCT02889978), we trained logistic regression classifiers under cross-validation to detect and localize cancer. Input features were methylation states from a targeted cfDNA assay of 2023 participants. TOO classification accuracy was 89% across 20 pre-specified prediction classes. We subsequently sought to understand the causes of the remaining TOO errors. 45% of the errors fell into clusters reflecting similarities in developmental biology, histology, or oncological drivers. Here, we analyzed tissues that may be affected by HPV-driven cancers; these accounted for 21% of TOO errors. The original classifier, which used only human epigenetic states as input, demonstrated cross-scoring between likely HPV-driven cancers of the anus (N = 14) and cervix (N = 11), as well as confirmed HPV-positive head & neck (H&N) cancers (37/62). We also observed HPV-associated vulva (N = 9) and penis (N = 1) cancers, which were not directly trained as TOO classes, were assigned high H&N scores. To test the hypothesis of HPV-driven TOO confusion, we assessed HPV cfDNA and HPV-driven methylation in human peripheral blood cfDNA, and used a specialist classifier restricted to HPV-associated cancers to resolve errors. We corroborated putative HPV-positive participants using targeted sequencing of HPV16 and HPV18 cfDNA fragments, and showed that the number of unique HPV-derived fragments in a sample matched with expected cancer localizations, HPV subtypes, and HPV status. Consistent with the literature, we found little evidence of HPV viremia in non-cancer participants despite the high prevalence of transient HPV infections in the US population. At 99.8% specificity, a cross-validated cutoff on the number of HPV cfDNA fragments in a sample achieved sensitivities of 78.6% (11/14), 36.3% (4/11), 66.6% (6/9), 100% (1/1), and 81.0% (30/37), for anus, cervix, vulva, penis, and confirmed HPV-positive H&N cancers, respectively. These sensitivities were similar to those achieved by the epigenetic classifier. Finally, we trained a cross-validated specialist classifier using the same features as the TOO classifier, but restricted to HPV-driven cancers. This improved TOO accuracy for detected anal cancers from 11% (1/9) to 100% (9/9), with little effect on other classes. These data support that HPV presence may explain observed cross-scoring patterns between H&N, cervix, and anus TOO prediction classes, which were driven by recurrent epigenomic changes in participants with HPV-positive cancers as detected by this assay. This suggests that modelling axes of shared biology across cancer types can be useful for accurate cfDNA TOO classification, which is critical to direct diagnostic work-up of diverse cancer types in a multi-cancer early detection test. Citation Format: Robert Calef, Oliver Venn, M. Cyrus Maher, John F. Beausang, Earl Hubbell, Aman Patel, Alexander P. Fields, Joerg Bredno, Arash Jamshidi, Alexander M. Aravanis. HPV-driven cancers show distinct methylation signatures in cell-free DNA (cfDNA) [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 2114.

Purpose To estimate the absolute number of adult survivors of childhood cancer in the U.S. population who carry a pathogenic or likely pathogenic variant in a cancer predisposition gene. Methods Using the Surveillance, Epidemiology, and End Results (SEER) Program, we estimated the number of childhood cancer survivors on December 31, 2016 for each childhood cancer diagnosis, multiplied this by the proportion of carriers of pathogenic/likely pathogenic variants in the St. Jude Lifetime Cohort (SJLIFE) study, and projected the resulting number onto the U.S. population. Results Based on genome sequence data, 11.8% of 2450 SJLIFE participants carry a pathogenic/likely pathogenic variant in one of 156 cancer predisposition genes. Given this information, we estimate that 21 800 adult survivors of childhood cancer in the United States carry a pathogenic/likely pathogenic variant in one of these genes. The highest estimated absolute number of variant carriers are among survivors of central nervous system tumors (n = 4300), particularly astrocytoma (n = 1800) and other gliomas (n = 1700), acute lymphoblastic leukemia (n = 4300), and retinoblastoma (n = 3500). The most frequently mutated genes are RB1 (n = 3000), NF1 (n = 2300), and BRCA2 (n = 800). Conclusion Given the increasing number of childhood cancer survivors in the United States, clinicians should counsel survivors regarding their potential genetic risk, consider referral for genetic counseling and testing, and, as appropriate, implement syndrome-specific cancer surveillance or risk-reducing measures.

Patients harboring germline pathogenic biallelic variants in genes involved in the recognition and repair of DNA damage are known to have a substantially increased cancer risk. Emerging evidence suggests that individuals harboring heterozygous variants in these same genes may also be at heightened, albeit lesser, risk for cancer. Herein, we sought to determine whether heterozygous variants in RECQL4, the gene encoding an essential DNA helicase that is defective in children with the autosomal recessive cancer-predisposing condition Rothmund–Thomson syndrome (RTS), are associated with increased risk for childhood cancer. To address this question, we interrogated germline sequence data from 4435 pediatric cancer patients at St. Jude Children's Research Hospital and 1127 from the National Cancer Institute Therapeutically Applicable Research to Generate Effective Treatment (TARGET) database and identified 24 (0.43%) who harbored loss-of-function (LOF) RECQL4 variants, including five of 249 (2.0%) with osteosarcoma (OS). These RECQL4 variants were significantly overrepresented in children with OS, the cancer most frequently observed in patients with RTS, as compared to 134,187 noncancer controls in the Genome Aggregation Database (gnomAD v2.1; P = 0.00087, odds ratio [OR] = 7.1, 95% CI, 2.9–17). Nine of the 24 (38%) individuals possessed the same c.1573delT (p.Cys525Alafs) variant located in the highly conserved DNA helicase domain, suggesting that disruption of this domain is central to oncogenesis. Altogether these data expand our understanding of the genetic factors predisposing to childhood cancer and reveal a novel association between heterozygous RECQL4 LOF variants and development of pediatric OS.

Variant interpretation in the era of massively parallel sequencing is challenging. Although many resources and guidelines are available to assist with this task, few integrated end-to-end tools exist. Here, we present the Pediatric Cancer Variant Pathogenicity Information Exchange (PeCanPIE), a web- and cloud-based platform for annotation, identification, and classification of variations in known or putative disease genes. Starting from a set of variants in variant call format (VCF), variants are annotated, ranked by putative pathogenicity, and presented for formal classification using a decision-support interface based on published guidelines from the American College of Medical Genetics and Genomics (ACMG). The system can accept files containing millions of variants and handle single-nucleotide variants (SNVs), simple insertions/deletions (indels), multiple-nucleotide variants (MNVs), and complex substitutions. PeCanPIE has been applied to classify variant pathogenicity in cancer predisposition genes in two large-scale investigations involving >4000 pediatric cancer patients and serves as a repository for the expert-reviewed results. PeCanPIE was originally developed for pediatric cancer but can be easily extended for use for nonpediatric cancers and noncancer genetic diseases. Although PeCanPIE's web-based interface was designed to be accessible to non-bioinformaticians, its back-end pipelines may also be run independently on the cloud, facilitating direct integration and broader adoption. PeCanPIE is publicly available and free for research use.

RECQL4 encodes an essential helicase that repairs DNA damage and maintains genomic stability. Biallelic pathogenic germline mutations in RECQL4 cause the autosomal recessive (AR) Rothmund-Thomson, Baller-Gerold, and RAPADILINO syndromes. Predisposition to cancer has been observed in all three syndromes, with osteosarcoma (OS) representing the greatest risk. Monoallelic pathogenic variants in DNA damage response genes (e.g., ATM, NBN ) are associated with a moderate increase in cancer risk. Building upon this notion, we sought to determine whether monoallelic RECQL4 loss of function (LOF) variants contribute to childhood cancer, particularly OS. Here, LOF variants were defined as nonsense, frameshift, or canonical splice altering variants that are classified as pathogenic or likely pathogenic based on the 2015 ACMG Guidelines. Among 4,436 pediatric cancer patients at St. Jude and 1,127 in the National Cancer Institute TARGET database (total: 5,563 patients), we identified 20 individuals (0.36%; tumor types in Table) harboring germline monoallelic RECQL4 LOF variants. Compared to reference controls in the Genome Aggregation Database (gnomAD), we observed an enrichment of RECQL4 LOF variants in pediatric cancer patients ( P = 0.046, prevalence ratio [PR] = 1.51). We next assessed for enrichment of RECQL4 LOF variants across tumor types (Table). This examination revealed a significant association between monoallelic RECQL4 mutations and leukemia ( P = 0.032, PR = 1.91) and more notably, with OS ( P = 0.0028, PR = 7.03) where 1.7% of pediatric OS patients carried monoallelic RECQL4 LOF variants. No evidence of association was observed for the other tumor types examined. Our data provide the first evidence of an association linking germline monoallelic RECQL4 LOF variants to childhood cancer, especially OS. Examination of larger cohorts are warranted to elucidate the extent to which these mutations increase the risk for leukemia and OS and the mechanisms by which they promote tumor formation. Citation Format: Jamie L. Maciaszek, Gang Wu, Kayla Hamilton, Rose B. McGee, Zhaoming Wang, Regina Nuccio, Stacy Hines-Dowell, Lynn Harrison, Elsie L. Gerhardt, Annastasia Ouma, Scott Newman, Aman Patel, Joy Nakitandwe, Elizabeth Azzato, Alberto S. Pappo, Sheila A. Shurtleff, David W. Ellison, James R. Downing, Melissa M. Hudson, Leslie L. Robison, Victor Santana, Jinghui Zhang, Kim E. Nichols, Chimene A. Kesserwan. Increased prevalence of germline monoallelic RECQL4 mutations in children with cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 3651.