Mark Gerstein

Quantum computing is gaining popularity due to its potential to detect complex patterns in data by leveraging unique quantum phenomena. It is particularly promising for complex data applications, such as those in biomedicine. However, in these contexts, it is often necessary to share and then aggregate data from multiple participants to achieve sufficient statistical power. Sharing data, especially sensitive information such as personal medical records, raises significant privacy concerns. To address these challenges, we propose a quantum-native method for encoding entire data cohorts directly into special quantum states, which we refer to as composite states. These quantum states provide a generic encoding suitable for a wide variety of downstream computations while preventing the inference of individual-level information. Quantum computations can be performed directly on composite states without accessing the underlying raw data. Building on this, we present the theoretical foundations of our scheme's utility and privacy guarantees, demonstrating resistance to membership inference attacks as measured via differential privacy. Furthermore, we introduce protocols that support multiparty collaborative quantum neural network training across diverse domains. Finally, we validate the effectiveness of composite states on three different datasets, focusing on genomic studies while also indicating how the approach can be applied in other domains without adaptation.

Diverse risk genes have been identified for neurodevelopmental disorders (NDDs), but how these genes converge on similar biological pathways in neurons, and thus give rise to similar phenotypes, is unclear. Here we apply a pooled CRISPR approach to successfully target 23 NDD loss-of-function genes with roles in chromatin biology and examine convergent effects on gene expression across human induced pluripotent stem cell-derived neural progenitor cells, glutamatergic neurons and GABAergic neurons. Points of convergence vary between these cell types, with the greatest number of convergent genes and strongest convergent networks in mature glutamatergic neurons, where they broadly represent synaptic, epigenetic and, unexpectedly, mitochondrial pathways. The most convergent networks were observed between NDD genes with shared biological annotations, clinical associations and co-expression patterns in human post-mortem brain. Drugs that were predicted to reverse convergent transcriptomic signatures and/or arousal and sensory processing behaviors ameliorated behavioral phenotypes in zebrafish NDD gene mutants. These results suggest that convergent effects of NDD risk genes could provide clinically useful insights.

Genomic variation between individuals is essential for understanding how differences in the genome sequence affect molecular and cellular processes. The Impact of Genomic Variation on Function (IGVF) Consortium aims to uncover the relationships among genomic variation, genome function, and phenotypes by combining experimental techniques, such as single-cell mapping and genomic perturbation assays, with computational approaches such as machine learning-based predictive modeling. The IGVF Data and Administrative Coordinating Centers collect, analyze, and disseminate data and results from across the consortium through an open-source platform called the IGVF Catalog. This resource includes, but is not limited to, data on the effects of coding variants on protein abundance and function, noncoding variants on enhancer activity (measured by MPRA or predicted computationally), and associations between variants and quantitative traits. All data are organized within a graph database comprising over 50 types of data collections with nearly 3 billion nodes and over 7.5 billion edges. The Catalog offers public API endpoints (https://api.catalogkg.igvf.org/) and a user-friendly interface for exploring, querying, and visualizing the data at https://catalog.igvf.org. We expect that this open-access platform will support the broader scientific community to advance our understanding of how genomic variation influences biology and disease.

Histological images offer a wealth of data. Mining these data holds significant potential for enhancing disease diagnosis and prognosis, though challenges remain, especially in noncancer contexts. In this study, we developed a statistical framework that links raw histological images and their derived features to the genotype, transcriptome, and chronological age of the samples. We first demonstrated an association between image features and genotypes, identifying 906 image quantitative trait loci (imageQTLs) significantly associated with image features. Next, we identified differentially expressed (DE) genes by stratifying samples into image-similar groups based on image features and performing DE comparisons between groups. Additionally, we developed a deep-learning model that accurately predicts gene expression in specific tissues from raw images and their features, highlighting gene sets associated with observed morphological changes. Finally, we constructed another deep-learning model to predict chronological age directly from raw images and their features, revealing relationships between age and tissue morphology, especially aspects derived from nucleus features. Both models are supported by a computational approach that greatly compresses gigapixel whole-slide images and extracts interpretable nucleus features, integrating both large-scale tissue morphology and smaller local structures. We have made all interpretable nucleus features, imageQTLs, DE genes, and deep-learning models available as online resources for further research.

The recent advent of long-read whole genome sequencing has enabled us to create an accurate telomere-to-telomere reference genome, construct pangenome graphs, and compile precise catalogs of genomic structural variations (SVs). These comprehensive SV repositories provide an excellent opportunity to explore the role of SVs in genotype-phenotype associations and examine the mechanisms by which SVs are introduced through double-strand break (DSB) repair. Here, we employed comprehensive SV catalogs identified through various short- and long-read whole genome sequencing efforts to infer the underlying mechanisms of SV introduction based on their genomic and epigenomic profiles. Our findings indicate that high local DNA methylation and DNA shape-related features, such as low variations in propeller twist, support the origins of homology-driven SVs. Subsequently, we utilized an active-learning-based unsupervised clustering approach, revealing that homology-dependent SVs show greater evidence of retaining ancestral recombination patterns compared to their homology-independent counterparts. Finally, our comparison of inherited and de novo SVs from healthy populations and rare disease cohorts showed distinct upstream H3K27me3 levels in de novo SVs from individuals with ultra-rare disorders. These findings highlight genome-wide characteristics that may influence the choice of repair mechanisms linked to heritable SV origins.

BACKGROUND: SINE variable number tandem repeat Alu elements (SVAs) are a unique group of hominid-specific composite retrotransposons with highly variable internal structure. They represent the youngest TE family in humans and contribute to genetic diversity, evolution, and disease. Recent findings indicate that SVA mobilization rates may exceed previous estimates, and many SVAs exhibit insertion polymorphism. SVAs facilitate transduction (TD) events when transcription initiates upstream of a source element, or when their internal termination signal is bypassed, mobilizing adjacent 5' and/or 3' sequence. To investigate features of non-reference SVA elements currently polymorphic in the human genome, we analyzed a structural variant callset built upon 35 diverse human genomes generated by the Human Genome Structural Variation Consortium. RESULTS: is a major contributor to SVA expansion in the human population. We further uncover that 40% of non-reference SVAs carry a TD on their 5' and/or 3' ends. Of these, the majority (69%) harbor sequence originating in a gene, including 14 exonic events and the mobilization of a processed pseudogene, supporting the role of SVA in exon shuffling. In addition, we identified a so-called "orphan" TD, defined by the absence of SVA sequence at the insertion site. Leveraging TD origin coordinates, we identify 55 active source elements, including nine non-reference and 46 across GRCh38 and T2T-CHM13, giving rise to 84% of TD-carrying SVAs. CONCLUSIONS: is more active than previously described and is a main driver of SVA expansion. We find two-fold more TD events compared to previous estimates, with an unexpected bias toward 3' events. Finally, we postulate that the discrepant SVA mobilization rate may be attributed to inter-individual variation in the presence/absence of source elements, a recent uptick in mobilization supported by overall low allele frequencies, and/or negative selection against deleterious insertions.

Background Hospital room sink drains contain biofilms that harbor opportunistic pathogens. Exposures to these pathogens may occur from aerosolization and droplet dispersion into patient rooms during sink use. This study characterizes aerosolization and droplet generation of sink drain opportunistic pathogens into operational hospital rooms. Methods Sink drains, sink surfaces, water droplets, aerosols generated during sink use, and settled aerosols were sampled in patient rooms and analyzed via culture-, spectrometry- and genome-based approaches. Opportunistic pathogens were compared across samples via whole-genome sequencing and single-nucleotide polymorphism analysis. Biofilms and settled aerosols underwent 16S ribosomal deoxyribonucleic acid sequencing to assess impacts of sink drain biofilms into room bioaerosols. Results Analyses suggested sink drain biofilm bacteria dispersed into hospital rooms. Opportunistic pathogens were identified in sink drains, droplets near sinks, and room aerosols. Stenotrophomonas maltophilia isolates from sink drain biofilm and droplets matched at the single-nucleotide level and microbial community analysis suggested general transmission of bacteria from sink drains into hospital rooms. Discussion Viable opportunistic pathogens from sink drains were present in water droplets and aerosols within patient range, suggesting a potential exposure route. Conclusions Hospital sink drain biofilms contributed to the microbiome of hospital room surfaces and air, with microbes generally transmitted from sink drain sources to the room.

Antimicrobial resistant pathogens and associated infections represent major public health threats affecting healthcare facilities, with sink drain biofilms serving as reservoirs for many of these bacteria. Despite attempts at sink drain biofilm disinfection and removal, drain biofilms inevitably regrow, and disinfection may shape the returning microbial communities and their resistance profiles. We applied culture-based and metagenomic approaches to study these drain disinfection effects on microbial community abundance, taxonomy, and antimicrobial resistance in operational hospital sinks. Drain biofilms regrew to baseline densities in approximately four days. Regrown biofilms contained more viable carbapenem-resistant bacteria and were dominated by Pseudomonadota, including Cupriavidus and Pseudomonas. Long-read sequencing revealed an increase in multidrug efflux pump genes after disinfection, which confer broad resistance to antibiotics and disinfectants. This work provides mechanistic insights into how disinfection influences sink drain biofilm ecology and the enrichment of antimicrobial resistance, with implications for infection prevention strategies in healthcare environments.

The Functional Annotation of Variants Online Resource (FAVOR), http://favor.genohub.org, is a whole genome variant annotation database and portal that provides comprehensive variant functional annotations of all possible variants across the genome. It can facilitate the analysis of whole-genome sequencing studies, support the interpretation of variant functional impacts, and help prioritize causal variants of diseases or traits. To support the growing popularity and expand the scope of FAVOR, we present here a substantial platform update. The new release features dramatically expanded annotations, a completely redesigned infrastructure powered by a newly implemented application programming interface (FAVOR-API), and a revamped web interface with advanced data-visualization capabilities and enhanced query performance. Key expansions include much more comprehensive variant annotations, including global, tissue- and cell-type-specific variant annotations; gene and protein annotations; support for both hg38 and hg19 reference genomes; and an interactive genome-browser for visualization of multi-faceted variant annotations. The updated platform also includes FAVOR-GPT, a large language model-powered interface for navigating the FAVOR database and interpreting results. FAVOR continues to evolve to keep pace with advances in research on interpreting the functional and phenotypic impact of genomic variation.

Pseudogenes have historically been regarded as nonfunctional remnants of genome evolution. However, relative to other noncoding genomic elements, their promoter architecture and epigenetic regulation remain incompletely understood. Here, we systematically characterize pseudogene promoters and compare them with those of protein-coding genes and long noncoding RNAs. To do this, we integrate matched transcriptomic and epigenomic data across 26 human tissues from the EN-TEx (ENCODE-GTEx) project. We uniformly annotate promoters with chromatin features (histone modifications, chromatin accessibility, and DNA methylation), sequence motifs, and evolutionary conservation, generating an online catalog. Leveraging this catalog, we show that, across multiple tissues, transcribed, unprocessed pseudogenes exhibit chromatin patterns similar to those of active protein-coding genes. In contrast, transcribed, processed pseudogenes show a different pattern: most lack the canonical hallmarks of transcription (e.g., active histone marks) at their promoters. Instead, their promoters show increased overlap with LINE elements, enrichment for YY1-like binding motifs, and higher Hi-C contact frequency, particularly with distal enhancer-like regulatory regions. Together with their greater conservation (relative to unprocessed pseudogenes), these features suggest that the transcription of processed pseudogenes may require regulatory mechanisms distinct from canonical promoter-associated epigenetic activation.

Neuropsychiatric disorders lack effective treatments due to a limited understanding of the underlying cellular and molecular mechanisms. To address this, we integrated population-scale single-cell genomics data and analyzed 23 cell-type-level gene regulatory networks across schizophrenia, bipolar disorder, and autism. Our analysis revealed potential druggable transcription factors co-regulating known risk genes that converge into cell-type-specific co-regulated modules. We applied graph neural networks on those modules to prioritize novel risk genes and leveraged them in a network-based drug repurposing framework to identify 220 drug molecules with the potential for targeting specific cell types. We found evidence for 37 of these drugs in reversing disorder-associated transcriptional phenotypes. Additionally, we discovered 335 drug-cell quantitative trait loci (eQTLs), revealing genetic variation's influence on drug target expression at the cell-type level. Our results provide a single-cell network medicine resource that provides potential mechanistic insights for advancing treatment options for neuropsychiatric disorders.

Abstract The recent advent of long-read whole genome sequencing has enabled us to create an accurate telomere-to-telomere reference genome, construct pangenome graphs, and compile precise catalogs of genomic structural variations (SVs). These comprehensive SV repositories provide an excellent opportunity to explore the role of SVs in genotype-phenotype associations and examine the mechanisms by which SVs are introduced through double-strand break (DSB) repair. Here, we employed comprehensive SV catalogs identified through various short- and long-read whole genome sequencing efforts to infer the underlying mechanisms of SV introduction based on their genomic and epigenomic profiles. Our findings indicate that high local DNA methylation and DNA shape-related features, such as low variations in propeller twist, support the origins of homology-driven SVs. Subsequently, we utilized an active-learning-based unsupervised clustering approach, revealing that the homology-dependent SVs show greater evidence of retaining ancestral recombination patterns compared to their homology-independent counterparts. Finally, our comparison of inherited and de novo SVs from healthy populations and rare disease cohorts showed distinct upstream H3K27me3 levels in de novo SVs from individuals with ultra-rare disorders. These findings highlight genome-wide characteristics that may influence the choice of repair mechanisms linked to heritable SV origins.

AI scientists powered by large language models have demonstrated substantial promise in autonomously conducting experiments and facilitating scientific discoveries across various disciplines. While their capabilities are promising, these agents also introduce novel vulnerabilities that require careful consideration for safety. However, there has been limited comprehensive exploration of these vulnerabilities. This perspective examines vulnerabilities in AI scientists, shedding light on potential risks associated with their misuse, and emphasizing the need for safety measures. We begin by providing an overview of the potential risks inherent to AI scientists, taking into account user intent, the specific scientific domain, and their potential impact on the external environment. Then, we explore the underlying causes of these vulnerabilities and provide a scoping review of the limited existing works. Based on our analysis, we propose a triadic framework involving human regulation, agent alignment, and an understanding of environmental feedback (agent regulation) to mitigate these identified risks. Furthermore, we highlight the limitations and challenges associated with safeguarding AI scientists and advocate for the development of improved models, robust benchmarks, and comprehensive regulations.

Quantum neural networks (QNNs) are gaining increasing interest due to their potential to detect complex patterns in data by leveraging uniquely quantum phenomena. This makes them particularly promising for biomedical applications. In these applications and in other contexts, increasing statistical power often requires aggregating data from multiple participants. However, sharing data, especially sensitive information like personal genomic sequences, raises significant privacy concerns. Quantum federated learning offers a way to collaboratively train QNN models without exposing private data. However, it faces major limitations, including high communication overhead and the need to retrain models when the task is modified. To overcome these challenges, we propose a privacy-preserving QNN training scheme that utilizes mixed quantum states to encode ensembles of data. This approach allows for the secure sharing of statistical information while safeguarding individual data points. QNNs can be trained directly on these mixed states, eliminating the need to access raw data. Building on this foundation, we introduce protocols supporting multi-party collaborative QNN training applicable across diverse domains. Our approach enables secure QNN training with only a single round of communication per participant, provides high training speed and offers task generality, i.e., new analyses can be conducted without reacquiring information from participants. We present the theoretical foundation of our scheme's utility and privacy protections, which prevent the recovery of individual data points and resist membership inference attacks as measured by differential privacy. We then validate its effectiveness on three different datasets with a focus on genomic studies with an indication of how it can used in other domains without adaptation.

Abstract Histological images offer a wealth of data. Mining these data holds significant potential for enhancing disease diagnosis and prognosis, though challenges remain, especially in non-cancer contexts. In this study, we developed a statistical framework that links raw histological images and their derived features to the genotype, transcriptome, and chronological age of the samples. We first demonstrated an association between image features and genotypes, identifying 906 image quantitative trait loci (imageQTLs) significantly associated with image features. Next, we identified differentially expressed (DE) genes by stratifying samples into image-similar groups based on image features and performing DE comparisons between groups. Additionally, we developed a deep-learning model that accurately predicts gene expression in specific tissues from raw images and their features, highlighting gene sets associated with observed morphological changes. Finally, we constructed another deep-learning model to predict chronological age directly from raw images and their features, revealing relationships between age and tissue morphology, especially aspects derived from nucleus features. Both models are supported by a computational approach that greatly compresses gigapixel whole-slide images and extracts interpretable nucleus features, integrating both large-scale tissue morphology and smaller local structures. We have made all interpretable nucleus features, imageQTLs, DE genes, and deep-learning models available as online resources for further research. Significance Statement This study establishes a comprehensive framework that links histological image features to genotype, transcriptome, and chronological age in large-scale healthy tissue datasets, providing valuable insights into tissue morphology. By identifying 906 significant, interpretable imageQTLs and conducting differential expression analysis based on image features, we enhance understanding of genetic and morphological interactions. Additionally, we developed predictive models for both gene expression and chronological age from raw histological images, introducing a novel approach to studying age-related tissue-specific changes and presenting the first model to demonstrate the predictability of age from histological images.

Abstract Although gene expression in the brain has been extensively investigated and compared to other tissues, the activity of pseudogenes has not been comprehensively surveyed. Here, leveraging large-scale RNA-seq data, we construct consistent pseudogene expression profiles in human and macaque brains and compare them to 29 other tissues. We further annotate pseudogenes with potential cellular roles based on co-clustering them with protein-coding genes. Notably, the majority of the expressed pseudogenes show elevated expression in the brain relative to other tissues, and these pseudogenes show broad and consistent expression patterns across brain subregions. Furthermore, spatiotemporal analyses reveal that pseudogenes in different brain subregions have greatly varying temporal trajectories (e.g., increasing vs. decreasing), in contrast to protein-coding genes that tend to be more uniform. Finally, we identify a set of pseudogenes exhibiting significant changes in neuropsychiatric disorders, some of which overlap with known brain eGenes (genes whose expression is associated with expression quantitative trait loci, eQTLs), as well as with genes implicated by genome-wide association studies (GWAS). Together, our study provides a public resource of pseudogene expression in human and macaque brains (in comparison to other tissues) and highlights aspects of pseudogene expression in brain development and pathogenesis.

In the version of the article initially published, the x -axis label for Fig. 4f was “Position (Mb)” and has now been corrected to “Position (kb)” in the HTML and PDF versions of the article.

Many genome-wide studies capture isolated moments in cell differentiation or organismal development. Conversely, longitudinal studies provide a more direct way to study these kinetic processes. Here, we present an approach for modeling gene-expression and chromatin kinetics from such studies: chronODE, an interpretable framework based on ordinary differential equations. chronODE incorporates two parameters that capture biophysical constraints governing the initial cooperativity and later saturation in gene expression. These parameters group genes into three major kinetic patterns: accelerators, switchers, and decelerators. Applying chronODE to bulk and single-cell time-series data from mouse brain development reveals that most genes (~87%) follow simple logistic kinetics. Among them, genes with rapid acceleration and high saturation values are rare, highlighting biochemical limitations that prevent cells from attaining both simultaneously. Early- and late-emerging cell types display distinct kinetic patterns, with essential genes ramping up faster. Extending chronODE to chromatin, we find that genes regulated by both enhancer and silencer cis-regulatory elements are enriched in brain-specific functions. Finally, we develop a bidirectional recurrent neural network to predict changes in gene expression from corresponding chromatin changes, successfully capturing the cumulative effect of multiple regulatory elements. Overall, our framework allows investigation of the kinetics of gene regulation in diverse biological systems.

Virtual cell modeling aims to predict cellular responses to diverse perturbations but faces challenges from biological complexity, multimodal data heterogeneity, and the need for interdisciplinary expertise. We introduce CellForge, a multi-agent framework that autonomously designs and synthesizes neural network architectures tailored to specific single-cell datasets and perturbation tasks. Given raw multi-omics data and task descriptions, CellForge discovers candidate architectures through collaborative reasoning among specialized agents, then generates executable implementations. Our core contribution is the framework itself: showing that multi-agent collaboration mechanisms - rather than manual human design or single-LLM prompting - can autonomously produce executable, high-quality computational methods. This approach goes beyond conventional hyperparameter tuning by enabling entirely new architectural components such as trajectory-aware encoders and perturbation diffusion modules to emerge from agentic deliberation. We evaluate CellForge on six datasets spanning gene knockouts, drug treatments, and cytokine stimulations across multiple modalities (scRNA-seq, scATAC-seq, CITE-seq). The results demonstrate that the models generated by CellForge are highly competitive with established baselines, while revealing systematic patterns of architectural innovation. CellForge highlights the scientific value of multi-agent frameworks: collaboration among specialized agents enables genuine methodological innovation and executable solutions that single agents or human experts cannot achieve. This represents a paradigm shift toward autonomous scientific method development in computational biology. Code is available at https://github.com/gersteinlab/CellForge.

Healthcare data are rapidly evolving with the introduction of new modalities and an exponential increase in volume. Current health data storage and communication services face major obstacles in terms of privacy, security, and operational efficiency in the face of this new data landscape. Blockchain technology, characterized by its immutability, auditability, and decentralization, is emerging as a promising solution. However, integrating blockchain into existing systems presents significant challenges, particularly regarding data privacy and scalability. This review aims to provide a comprehensive understanding of how blockchain technology can transform the biomedical sector, potentially making healthcare data management more secure and efficient.

Abstract Diverse sets of complete human genomes are required to construct a pangenome reference and to understand the extent of complex structural variation. Here we sequence 65 diverse human genomes and build 130 haplotype-resolved assemblies (median continuity of 130 Mb), closing 92% of all previous assembly gaps 1,2 and reaching telomere-to-telomere status for 39% of the chromosomes. We highlight complete sequence continuity of complex loci, including the major histocompatibility complex (MHC), SMN1 / SMN2 , NBPF8 and AMY1/AMY2 , and fully resolve 1,852 complex structural variants. In addition, we completely assemble and validate 1,246 human centromeres. We find up to 30-fold variation in α-satellite higher-order repeat array length and characterize the pattern of mobile element insertions into α-satellite higher-order repeat arrays. Although most centromeres predict a single site of kinetochore attachment, epigenetic analysis suggests the presence of two hypomethylated regions for 7% of centromeres. Combining our data with the draft pangenome reference 1 significantly enhances genotyping accuracy from short-read data, enabling whole-genome inference 3 to a median quality value of 45. Using this approach, 26,115 structural variants per individual are detected, substantially increasing the number of structural variants now amenable to downstream disease association studies.

These data tables contain results and statistical analyses from the Seahorse Mito Stress assay and TOMM20 immunostaining for mitochondrial morphology. Detailed methods and final figures are available at DOI: https://doi.org/10.1101/2024.08.23.609190

These data tables contain results and statistical analyses from the Seahorse Mito Stress assay and TOMM20 immunostaining for mitochondrial morphology. Detailed methods and final figures are available at DOI: https://doi.org/10.1101/2024.08.23.609190

Genome-wide association studies (GWASs) and expression analyses implicate noncoding regulatory regions as harboring risk factors for psychiatric disease, but functional characterization of these regions remains limited. Here, we perform capture STARR-sequencing of over 70,000 candidate regions to identify active enhancers in primary human neural progenitor cells (phNPCs). We select candidate regions by integrating data from NPCs, prefrontal cortex, developmental timepoints, and GWASs. Over 8000 regions demonstrate enhancer activity in the phNPCs, and we link these regions to over 2200 predicted target genes. These genes are involved in neuronal and psychiatric disease-associated pathways, including neuronal system, nervous system development, and developmental delay. We functionally validate a subset of these enhancers using mutation STARR-sequencing and CRISPR deletions, demonstrating the effects of genetic variation on enhancer activity and enhancer deletion on gene expression. Overall, we identify thousands of highly active enhancers and functionally validated a subset of these enhancers, improving our understanding of regulatory networks underlying brain function and disease.

Non-canonical (i.e., unannotated) open reading frames (ncORFs) have until recently been omitted from reference genome annotations, despite evidence of their translation, limiting their incorporation into biomedical research. To address this, in 2022, we initiated the TransCODE consortium and built the first community-driven consensus catalog of human ncORFs, which was openly distributed to the research community via Ensembl-GENCODE. While this catalog represented a starting point for reference ncORF annotation, major technical and scientific issues remained. In particular, this initial catalogue had no standardized framework to judge the evidence of translation for individual ncORFs. Here, we present an expanded and refined catalog of the human reference annotation of ncORFs. By incorporating more datasets and by lifting constraints on ORF length and start-codon, we define a comprehensive set of 28,359 ncORFs that is nearly four times the size of the previous catalog. Furthermore, to aid users who wish to work with ncORFs with the strongest and most reproducible signals of translation, we utilized a data-driven framework (i.e. translation signature scores) to assess the accumulated evidence for any individual ncORF. Using this approach, we derive a subset of 7,888 ncORFs with translation evidence on par with canonical protein-coding genes, which we refer to as the Primary set. This set can serve as a reliable reference for downstream analyses and validation, with a particular emphasis on high quality. Overall, this update reflects continual community-driven efforts to make ncORFs accessible and actionable to the broader research public and further iterations of the catalog will continue to expand and refine this resource.

Background Epigenetic modifications play a vital role in the pathogenesis of human diseases, particularly neurodegenerative disorders such as Alzheimer's disease (AD), where dysregulated histone modifications are strongly implicated in disease mechanisms. While recent advances underscore the importance of accurately identifying these modifications to elucidate their contribution to AD pathology, existing computational methods remain limited by their generic approaches that overlook disease-specific epigenetic signatures. Results To bridge this gap, we developed a novel large language model (LLM)-based deep learning framework tailored for disease-contextual prediction of histone modifications and variant effects. Focusing on AD as a case study, we integrated epigenomic data from multiple patient samples to construct a comprehensive, disease-specific histone modification dataset, enabling our model to learn AD-associated molecular signatures. A key innovation of our approach is the incorporation of a Mixture of Experts (MoE) architecture, which effectively distinguishes between disease and healthy epigenetic states, allowing for precise identification of AD-relevant epigenetic modification patterns. Our model demonstrates robust performance in disease-specific histone modification prediction, achieving mean area under receiver-operating curves (AUROCs) ranging from 0.7863 to 0.9142, significantly outperforming existing state-of-the-art methods that lack disease context. Beyond accurate modification site prediction, our framework provides important biological insights by successfully prioritizing AD-associated genetic variants, which show significant enrichment in disease-relevant pathways, supporting their potential functional role in AD pathogenesis. These findings suggest that differential modification loci identified by our model may represent key regulatory elements in AD. Conclusions Our framework establishes a powerful new paradigm for epigenetic research that can be extended to other complex diseases, offering both a valuable tool for variant effect interpretation and a promising strategy for uncovering novel disease mechanisms through epigenetic profiling.

Large language models have shown promise in clinical decision making, but current approaches struggle to localize and correct errors at specific steps of the reasoning process. This limitation is critical in medicine, where identifying and addressing reasoning errors is essential for accurate diagnosis and effective patient care. We introduce Med-PRM, a process reward modeling framework that leverages retrieval-augmented generation to verify each reasoning step against established medical knowledge bases. By verifying intermediate reasoning steps with evidence retrieved from clinical guidelines and literature, our model can precisely assess the reasoning quality in a fine-grained manner. Evaluations on five medical QA benchmarks and two open-ended diagnostic tasks demonstrate that Med-PRM achieves state-of-the-art performance, with improving the performance of base models by up to 13.50% using Med-PRM. Moreover, we demonstrate the generality of Med-PRM by integrating it in a plug-and-play fashion with strong policy models such as Meerkat, achieving over 80\% accuracy on MedQA for the first time using small-scale models of 8 billion parameters. Our code and data are available at: https://med-prm.github.io/

Spatially resolved in vivo CRISPR screening integrates gene editing with spatial transcriptomics to examine how genetic perturbations alter gene expression within native tissue environments. However, current methods are limited to small perturbation panels and the detection of a narrow subset of protein-coding RNAs. We present Perturb-DBiT, a distinct and versatile approach for the simultaneous co-sequencing of spatial total RNA whole-transcriptome and single-guide RNAs (sgRNAs), base-by-base, on the same tissue section. This method enables unbiased discovery of how genetic perturbations influence RNA regulation, cellular dynamics, and tissue architecture in situ. Applying Perturb-DBiT to a human cancer metastatic colonization model, we mapped large panels of sgRNAs across tumor colonies in consecutive tissue sections alongside their corresponding total RNA transcriptomes. This revealed novel insights into how perturbations affect long non-coding RNA (lncRNA) co-variation, microRNA–mRNA interactions, and global and distinct tRNA alterations in amino acid metabolism linked to tumor migration and growth. By integrating transcriptional pseudotime trajectories, we further uncovered the impact of perturbations on clonal dynamics and cooperation. In an immune-competent syngeneic mouse model, Perturb-DBiT enabled investigation of genetic perturbations within the tumor immune microenvironment, revealing distinct and synergistic effects on immune infiltration and suppression. Perturb-DBiT provides a spatially resolved comprehensive view of how genetic knockouts influence diverse molecular and cellular responses including small and large RNA regulation, tumor proliferation, migration, metastasis, and immune interactions, offering a panoramic perspective on perturbation responses in complex tissues.

Large language models (LLMs) and large multimodal models (LMMs) promise to accelerate biomedical discovery, yet their reliability remains unclear. We introduce ARIEL (AI Research Assistant for Expert-in-the-Loop Learning), an open-source evaluation and optimization framework that pairs a curated multimodal biomedical corpus with expert-vetted tasks to probe two capabilities: full-length article summarization and fine-grained figure interpretation. Using uniform protocols and blinded PhD-level evaluation, we find that state-of-the-art models generate fluent but incomplete summaries, whereas LMMs struggle with detailed visual reasoning. We later observe that prompt engineering and lightweight fine-tuning substantially improve textual coverage, and a compute-scaled inference strategy enhances visual question answering. We build an ARIEL agent that integrates textual and visual cues, and we show it can propose testable mechanistic hypotheses. ARIEL delineates current strengths and limitations of foundation models, and provides a reproducible platform for advancing trustworthy AI in biomedicine.

A major challenge in near-term quantum computing is its application to large real-world datasets due to scarce quantum hardware resources. One approach to enabling tractable quantum models for such datasets involves finding low-dimensional representations that preserve essential information for downstream analysis. In classical machine learning, variational autoencoders (VAEs) facilitate efficient data compression, representation learning for subsequent tasks, and novel data generation. However, no quantum model has been proposed that exactly captures all of these features for direct application to quantum data on quantum computers. Some existing quantum models for data compression lack regularization of latent representations, thus preventing direct use for generation and control of generalization. Others are hybrid models with only some internal quantum components, impeding direct training on quantum data. To address this, we present a fully quantum framework, $\ensuremath{\zeta}$-QVAE, which encompasses all the capabilities of classical VAEs and can be directly applied to map both classical and quantum data to a lower-dimensional space, while effectively reconstructing much of the original state from it. Our model utilizes regularized mixed states to attain optimal latent representations. It accommodates various divergences for reconstruction and regularization. Furthermore, by accommodating mixed states at every stage, it can utilize the full data density matrix and allow for a training objective defined on probabilistic mixtures of input data. Doing so, in turn, makes efficient optimization possible and has potential implications for private and federated learning. In addition to exploring the theoretical properties of $\ensuremath{\zeta}$-QVAE, we demonstrate its performance on representative genomics and synthetic data. Our results indicate that $\ensuremath{\zeta}$-QVAE consistently learns representations that better utilize the capacity of the latent space and exhibits similar or better performance compared with matched classical models.

Background College students are vulnerable to setting long-term trajectories of low physical activity (PA) but are reachable via mobile health fitness tracking (eg, mobile health step counting) and interpersonal support tailored to the college community. However, no studies have statistically isolated the appeal and influence of these intra- and interpersonal components in college-based PA interventions. Objective This study retrospectively examined a college-based PA promotion program at a northeast US public university during the COVID-19 pandemic to (1) test the impact of student status on the use of intervention components and (2) determine whether such use was associated with successful retention and goal achievement in the program. Methods The university used a commercial platform for a 30-day PA promotion program during April 2021 with intrapersonal (step-tracker syncing, education, self-monitoring, and motivational messaging) and interpersonal (friend interactions and team games) components. App use was operationalized as intrapersonal (frequency of opening app, education, and self-monitoring) and interpersonal (friends made in-app and team affiliation and size). Results Campus-wide emails elicited sign-up by 156 undergraduate students, 57 graduate students, and 126 faculty and staff members. Objective 1 yielded the following results: undergraduates used the app less frequently (median 0.8, IQR 0.4-1.7 times per day) than other groups (graduate students: median 1.4, IQR 0.7-2.7 times per day; P=.01; faculty: median 1.3, IQR 0.7-2.7 times per day, H2=14.5; P=.001) but made the same number of friends (median 1-2) and teammates (median 8-9; P=.77 for friends and P=.93 for teammates). Objective 2 yielded the following results: most participants (313/335, 93.4%; 95% CI 90%-96%) were retained for the first 7 days, but by 30 days, retention dropped, most notably for undergraduate students (82/154, 53.2%; 95% CI 45%-61%), followed by graduate students (39/56, 70%; 95% CI 56%-81%) and faculty and staff (93/125, 74.4%; 95% CI 66%-82%; χ22=12.6; P<.001). Retention was associated with app engagement frequency (model hazard ratio 0.56, 95% CI 0.43-0.72; P<.001) and affiliation with a team having high median app engagement and a large size (intracluster correlation coefficient 0.064, 95% CI 0.001-0.164, P=.05). Meeting a daily step goal was associated with app engagement frequency (β=.72, SE=0.21; P=.001), number of friends (β=.40, SE 0.20; P=.04), and an initial motive of maintaining or increasing (rather than starting) PA (β=.99, SE=0.21; P<.001). Conclusions College students, compared with faculty and staff, used the app less frequently, used the app for a shorter duration before abandonment, and met the step goal on fewer days. Engagement with the program was associated with longer retention and better PA outcomes, which were critically modified by the interpersonal engagement. These findings suggest that college students using virtual PA support during times of physical isolation could benefit from more tailored implementation strategies (eg, timed prompts and team reassignments).

Blockchain technology is gaining traction in the biomedical sector due to its ability to improve trust and reduce the risk of fraud and errors in health data management. However, the large volume of biomedical datasets has slowed its adoption due to poor scalability. This challenge is especially relevant for applications that rely on blockchain's strong immutability by storing data directly on-chain. In this work, we demonstrate the potential of blockchain to create a secure and trustless environment for managing large on-chain records. Specifically, we detail an efficient, index-based approach for storing data on the Ethereum blockchain. We show that insertion and retrieval speeds remain nearly constant relative to database size, scaling linearly with the amount of data processed. Additionally, we achieve substantial efficiency gains through low-level assembly optimizations on the Ethereum Virtual Machine, highlighting the limitations of the Solidity compiler. Finally, we illustrate this approach through a practical case study, by designing and implementing a smart contract for storing and querying training certificates on the Ethereum blockchain. Our solution achieves 2x faster data insertion, 500x faster retrieval, 60% lower gas costs, and 50% lower storage usage compared to baseline methods. It won first place for track 1 of the 2022 iDASH secure genome analysis competition. We also demonstrate that this solution readily adapts to other data types, enabling efficient on-chain storage and retrieval of text, RNA-seq, or biomedical image data.

We introduce a Discard-and-Restart molecular dynamics (MD) algorithm tailored for the sampling of realistic protein intermediate states. It aids computational structure-based drug discovery by reducing the simulation times to compute a "quick sketch" of folding pathways by up to 2000x. The algorithm iteratively performs short MD simulations and measures their proximity to a target state via a collective variable (CV) loss, which can be defined in a flexible fashion, locally or globally. Using the loss, if the trajectory proceeds toward the target, the MD simulation continues. Otherwise, it is discarded, and a new MD simulation is restarted, with new initial velocities randomly drawn from a Maxwell-Boltzmann distribution. The discard-and-restart algorithm demonstrates efficacy and atomistic accuracy in capturing the folding pathways in several contexts: (1) fast-folding small protein domains; (2) the folding intermediate of the prion protein PrP; and (3) the spontaneous partial unfolding of α-Tubulin, a crucial event for microtubule severing. During each iteration of the algorithm, we can perform AI-based analysis of the transitory conformations to find potential binding pockets, which could represent druggable sites. Overall, our algorithm enables systematic and computationally efficient exploration of conformational landscapes, enhancing the design of ligands targeting dynamic protein states.

A statistician offers insight into risk, probability, and decision-making

Complex medical reasoning requires integrating heterogeneous clinical evidence across multiple inference steps. Large language models (LLMs) now approach this through two routes: internalized reasoning and externalized agent scaffolding (frameworks that decompose problems collaboratively amongst multiple LLMs). To determine whether these routes are exclusive or complementary, we introduce MedicalAgentsBench, a filtered benchmark of 862 complex clinical questions drawn from the union of eight medical datasets via difficulty-aware curation and contamination screening. Evaluating three internalized reasoning models (DeepSeek-R1, o1-mini, and o3-mini), seven base models, and nine externalized agent-based methods, we find that internalized and externalized approaches each independently improve performance, and that their benefits compound: the highest accuracy is achieved by layering agent workflows onto an internalized reasoning model (i.e., o3-mini + MDAgents with 35.1%). Pareto analysis shows this combination dominates the cost-performance frontier; moreover, lightweight optimization on inexpensive models offers an entry point for resource-constrained settings. Our benchmark is at https://github.com/gersteinlab/MedicalAgentsBench.

Density Functional Theory (DFT) is a pivotal method within quantum chemistry and materials science, with its core involving the construction and solution of the Kohn-Sham Hamiltonian. Despite its importance, the application of DFT is frequently limited by the substantial computational resources required to construct the Kohn-Sham Hamiltonian. In response to these limitations, current research has employed deep-learning models to efficiently predict molecular and solid Hamiltonians, with roto-translational symmetries encoded in their neural networks. However, the scalability of prior models may be problematic when applied to large molecules, resulting in non-physical predictions of ground-state properties. In this study, we generate a substantially larger training set (PubChemQH) than used previously and use it to create a scalable model for DFT calculations with physical accuracy. For our model, we introduce a loss function derived from physical principles, which we call Wavefunction Alignment Loss (WALoss). WALoss involves performing a basis change on the predicted Hamiltonian to align it with the observed one; thus, the resulting differences can serve as a surrogate for orbital energy differences, allowing models to make better predictions for molecular orbitals and total energies than previously possible. WALoss also substantially accelerates self-consistent-field (SCF) DFT calculations. Here, we show it achieves a reduction in total energy prediction error by a factor of 1347 and an SCF calculation speed-up by a factor of 18%. These substantial improvements set new benchmarks for achieving accurate and applicable predictions in larger molecular systems.

Large language models (LLMs) have dramatically enhanced the field of language intelligence, as demonstrably evidenced by their formidable empirical performance across a spectrum of complex reasoning tasks. Additionally, theoretical proofs have illuminated their emergent reasoning capabilities, providing a compelling showcase of their advanced cognitive abilities in linguistic contexts. Critical to their remarkable efficacy in handling complex reasoning tasks, LLMs leverage the intriguing chain-of-thought (CoT) reasoning techniques, obliging them to formulate intermediate steps en route to deriving an answer. The CoT reasoning approach has not only exhibited proficiency in amplifying reasoning performance but also in enhancing interpretability, controllability, and flexibility. In light of these merits, recent research endeavors have extended CoT reasoning methodologies to nurture the development of autonomous language agents, which adeptly adhere to language instructions and execute actions within varied environments. This survey article orchestrates a thorough discourse, penetrating vital research dimensions, encompassing (i) the foundational mechanics of CoT techniques, with a focus on elucidating the circumstances and justification behind its efficacy; (ii) the paradigm shift in CoT; and (iii) the burgeoning of language agents fortified by CoT approaches. Prospective research avenues envelop explorations into generalization, efficiency, customization, scaling, and safety. A repository for the related papers is available at https://github.com/Zoeyyao27/CoT-Igniting-Agent .