Yuri Alexeev, Maximilian Amsler, Marco Antonio Barroca, Sanzio Bassini, Torey Battelle, Daan Camps, David Casanova, Young Jay Choi, Frederic T. Chong, Charles Chung, Christopher Codella, Antonio D. Córcoles, James Cruise, Alberto Di Meglio, Ivan Duran, Thomas Eckl, Sophia Economou, Stephan Eidenbenz, Bruce Elmegreen, Clyde Fare, Ismael Faro, Cristina Sanz Fernández, Rodrigo Neumann Barros Ferreira, Keisuke Fuji, Bryce Fuller, Laura Gagliardi, Giulia Galli, Jennifer R. Glick, Isacco Gobbi, Pranav Gokhale, Salvador de la Puente Gonzalez, Johannes Greiner, Bill Gropp, Michele Grossi, Emanuel Gull, Burns Healy, Matthew R. Hermes, Benchen Huang, Travis S. Humble, Nobuyasu Ito, Artur F. Izmaylov, Ali Javadi-Abhari, Douglas Jennewein, Shantenu Jha, Liang Jiang, Barbara Jones, Wibe Albert de Jong, Petar Jurcevic, William Kirby, Stefan Kister, Masahiro Kitagawa, Joel Klassen, Katherine Klymko, Kwangwon Koh, Masaaki Kondo, Dog̃a Murat Kürkçüog̃lu, Krzysztof Kurowski, Teodoro Laino, Ryan Landfield, Matt Leininger, Vicente Leyton-Ortega, Ang Li, Meifeng Lin, Junyu Liu, Nicolas Lorente, Andre Luckow, Simon Martiel, Francisco Martin-Fernandez, Margaret Martonosi, Claire Marvinney, Arcesio Castaneda Medina, Dirk Merten, Antonio Mezzacapo, Kristel Michielsen, Abhishek Mitra, Tushar Mittal, Kyungsun Moon, Joel Moore, Sarah Mostame, Mario Motta, Young-Hye Na, Yunseong Nam, Prineha Narang, Yu-ya Ohnishi, Daniele Ottaviani, Matthew Otten, Scott Pakin, Vincent R. Pascuzzi, Edwin Pednault, Tomasz Piontek, Jed Pitera, Patrick Rall, Gokul Subramanian Ravi, Niall Robertson, Matteo A.C. Rossi, Piotr Rydlichowski, Hoon Ryu, Georgy Samsonidze, Mitsuhisa Sato, Nishant Saurabh, Vidushi Sharma, Kunal Sharma, Soyoung Shin, George Slessman, Mathias Steiner, Iskandar Sitdikov, In-Saeng Suh, Eric D. Switzer, Wei Tang, Joel Thompson, Synge Todo, Minh C. Tran, Dimitar Trenev, Christian Trott, Huan-Hsin Tseng, Norm M. Tubman, Esin Tureci, David García Valiñas, Sofia Vallecorsa, Christopher Wever, Konrad Wojciechowski, Xiaodi Wu, Shinjae Yoo, Nobuyuki Yoshioka, Victor Wen-zhe Yu, Seiji Yunoki, Sergiy Zhuk, Dmitry Zubarev <<<

David S. Tsao, Sukrit Silas, Brian P. Landry, Nelda P. Itzep, Amy B. Nguyen, Samuel Greenberg, Celeste K. Kanne, Vivien A. Sheehan, Rani Sharma, Rahul Shukla, Prem N. Arora, Oguzhan Atay <<<

Abstract Next-generation DNA sequencing is currently limited by an inability to accurately count the number of input DNA molecules. Molecular counting is particularly needed when accurate quantification is required for diagnostic purposes, such as in single gene non-invasive prenatal testing (sgNIPT) and liquid biopsy. We developed Quantitative Counting Template (QCT) molecular counting to reconstruct the number of input DNA molecules using sequencing data. We then used QCT molecular counting to develop sgNIPTs of sickle cell disease, cystic fibrosis, spinal muscular atrophy, alpha-thalassemia, and beta-thalassemia. The analytical sensitivity and specificity of sgNIPT was >98% and >99%, respectively. Validation of sgNIPTs was further performed with maternal blood samples collected during pregnancy, and sgNIPTs were 100% concordant with newborn follow-up. <<<

Abstract Stem-like progenitors are self-renewing cytotoxic T cells that expand as effector cells during successful checkpoint immunotherapy 1,2 . Emerging evidence suggests that tumour-draining lymph nodes support the continuous generation of these stem-like cells that replenish tumour sites and are a key source of expanded effector populations 3–6 , underlining the importance of understanding what factors promote and maintain activated T cells in the stem-like state. Here, using advanced three-dimensional multiplex immunofluorescence imaging, we identify antigen-presentation niches in tumour-draining lymph nodes that support the expansion, maintenance and affinity evolution of TCF-1 + PD-1 + SLAMF6 high stem-like CD8 + T cells. Contrary to the prevailing view that persistent T cell receptor (TCR) signalling drives terminal effector differentiation, prolonged antigen engagement days beyond initial priming sustains the proliferation and self-renewal of these stem-like T cells in vivo. The inhibitory PD-1 pathway has a central role in this process through fine-tuning the TCR signal input that enables the selective expansion of high-affinity TCR stem-like clones as a renewable source of effector cells. PD-1 blockade disrupts this tuning, leading to terminal differentiation or death of the most avid anti-tumour stem-like cells. Our results therefore reveal a relationship between TCR ligand affinity recognition, a key negative-feedback regulatory loop and T cell stemness programming. Furthermore, these findings raise questions about whether anti-PD-1 blockade during cancer immunotherapy provides a short-term anti-tumour effect at the cost of diminishing efficacy due to progressive loss of these critical high-affinity precursors. <<<

The success of today's large language models (LLMs) depends on the observation that larger models perform better. However, the origin of this neural scaling law, that loss decreases as a power law with model size, remains unclear. We propose that representation superposition, meaning that LLMs represent more features than they have dimensions, can be a key contributor to loss and cause neural scaling. Based on Anthropic's toy model, we use weight decay to control the degree of superposition, allowing us to systematically study how loss scales with model size. When superposition is weak, the loss follows a power law only if data feature frequencies are power-law distributed. In contrast, under strong superposition, the loss generically scales inversely with model dimension across a broad class of frequency distributions, due to geometric overlaps between representation vectors. We confirmed that open-sourced LLMs operate in the strong superposition regime and have loss scaling inversely with model dimension, and that the Chinchilla scaling laws are also consistent with this behavior. Our results identify representation superposition as a central driver of neural scaling laws, providing insights into questions like when neural scaling laws can be improved and when they will break down. <<<

The dominant sequence transduction models are based on complex recurrent or convolutional neural networks in an encoder-decoder configuration. The best performing models also connect the encoder and decoder through an attention mechanism. We propose a new simple network architecture, the Transformer, based solely on attention mechanisms, dispensing with recurrence and convolutions entirely. Experiments on two machine translation tasks show these models to be superior in quality while being more parallelizable and requiring significantly less time to train. Our model achieves 28.4 BLEU on the WMT 2014 English-to-German translation task, improving over the existing best results, including ensembles by over 2 BLEU. On the WMT 2014 English-to-French translation task, our model establishes a new single-model state-of-the-art BLEU score of 41.8 after training for 3.5 days on eight GPUs, a small fraction of the training costs of the best models from the literature. We show that the Transformer generalizes well to other tasks by applying it successfully to English constituency parsing both with large and limited training data. <<<

Abstract Embedding is the key step in single-cell Hi-C (scHi-C) analysis which relies on capturing biological meaningful heterogeneity at various levels of genome architecture. To understand the strength and limitations of existing tools in various applications, here we use ten scHi-C datasets to benchmark thirteen embedding tools including Va3DE, a new convolutional neural network model that can accommodate large cell numbers. We built a software framework to decouple the preprocessing options of existing tools and found that no single tool works best across all datasets under default settings. The difficulty levels and preferred resolutions are different between benchmark datasets, and the choice of data representation and preprocessing strongly impact the embedding performance. Embedding cells from early embryonic stages relies on long-range compartment-scale contacts, but resolving cell cycle phases and complex tissue requires short-range loop-scale contacts. Both random-walk and inverse document frequency (IDF) transformation prefers long-range “compartment-scale” over short-range “loop-scale” embedding, while deep-learning methods better overcome sparsity at both scales and are more versatile with different resolutions. Finally, “diagonal integration” with independent data modal is a promising approach to distinguish similar cell subpopulations. Our findings underscore the significance of appropriate priors for scHi-C embedding and also offer insights into genome architecture heterogeneity. <<<

Haitang Liao, Ruiyan Ma, Shuai Hao, Xuxin Tan, Xue Zeng, Rui Song, Bing Chen, Zhezhe Cao, Wei Shen, Zhenchun Luo, Jianfeng Huang, He Huang, Liangming Liu, Chenyang Duan <<<

Abstract Accurate early diagnosis and prognosis of sepsis remain major clinical challenges. This study explores specific plasma cell‐free mitochondrial DNA (cfmtDNA) breakage patterns as potential biomarkers for sepsis. Plasma samples from ten non‐sepsis control patients and 63 sepsis patients are analyzed using mitochondrial DNA fragmentomics, revealing distinct breakage sites in the RNR2 (positions 2474–2478) and COX2 (positions 7761, 7775, 7776, 7777, and 7783) regions, which are intact in healthy individuals but exhibited high‐frequency breakage in sepsis patients. Diagnostic models based on these breakage sites show superior accuracy for sepsis detection (AUC = 0.865) and prognosis prediction (AUC = 0.809) compared to traditional cfmtDNA copy number assessments. Notably, COX2 breakage frequency correlated with inflammatory markers and SOFA scores, highlighting its prognostic potential. Mechanistic analyses suggest that reduced protein binding in sepsis may increase cfmtDNA susceptibility to cleavage by bacterial restriction endonucleases. These findings indicate that plasma cfmtDNA breakage characteristics can serve as valuable biomarkers for early sepsis detection and therapeutic monitoring. <<<

Mingfei Chen, Yifan Wang, Zhengqin Li, Homanga Bharadhwaj, Yujin Chen, Chuan Qin, Ziyi Kou, Yuan Tian, Eric Whitmire, Rajinder Sodhi, Hrvoje Benko, Eli Shlizerman, Yue Liu <<<

Prior works on 3D hand trajectory prediction are constrained by datasets that decouple motion from semantic supervision and by models that weakly link reasoning and action. To address these, we first present the EgoMAN dataset, a large-scale egocentric dataset for interaction stage-aware 3D hand trajectory prediction with 219K 6DoF trajectories and 3M structured QA pairs for semantic, spatial, and motion reasoning. We then introduce the EgoMAN model, a reasoning-to-motion framework that links vision-language reasoning and motion generation via a trajectory-token interface. Trained progressively to align reasoning with motion dynamics, our approach yields accurate and stage-aware trajectories with generalization across real-world scenes. <<<

Science is characterized by collaboration and cooperation, but also by uncertainty, competition, and inequality. While there has always been some concern that these pressures may compel some to defect from the scientific research ethos—i.e., fail to make genuine contributions to the production of knowledge or to the training of an expert workforce—the focus has largely been on the actions of lone individuals. Recently, however, reports of coordinated scientific fraud activities have increased. Some suggest that the ease of communication provided by the internet and open-access publishing have created the conditions for the emergence of entities—paper mills (i.e., sellers of mass-produced low quality and fabricated research), brokers (i.e., conduits between producers and publishers of fraudulent research), predatory journals, who do not conduct any quality controls on submissions—that facilitate systematic scientific fraud. Here, we demonstrate through case studies that i) individuals have cooperated to publish papers that were eventually retracted in a number of journals, ii) brokers have enabled publication in targeted journals at scale, and iii), within a field of science, not all subfields are equally targeted for scientific fraud. Our results reveal some of the strategies that enable the entities promoting scientific fraud to evade interventions. Our final analysis suggests that this ability to evade interventions is enabling the number of fraudulent publications to grow at a rate far outpacing that of legitimate science. <<<

Aeron M. Small, Ta-Yu Yang, Shinsuke Itoh, Sébastien Thériault, Line Dufresne, Ryo Kurosawa, Issei Komuro, Koichi Matsuda, Ha My T. Vy, Eric H. Farber-Eger, Lauren Lee Shaffer, Kristin M. Boulier, Kristin M. Corey, Megan E. Ramaker, Fabien Laporte, Jean-Jacques Schott, Solena Le Scouarnec, Sasha A. Singh, Abhijeet R. Sonawane, Harry A. Smith, Nicholas Rafaels, , Jonas Ghouse, Anna A. Raja, Sisse R. Ostrowski, Erik Sørensen, Christina Mikkelsen, Ole B. Pedersen, Christian Erikstrup, Henrik Ullum, , Gardar Sveinbjornsson, Daniel F. Gudbjartsson, Erik Abner, , Jiwoo Lee, Andrea Ganna, Ulrike Nowak-Göttl, Sarah Finer, , David A. van Heel, Johannes Schumacher, Carlo Maj, Baravan Al-Kassou, Georg Nickenig, Teresa Trenkwalder, Martina Dreβen, Markus Krane, Markus M. Nöthen, Marta R. Moksnes, Ben M. Brumpton, Stacey Knight, Kirk U. Knowlton, Lincoln Nadauld, Radek Debiec, Muntaser D. Musameh, Peter S. Braund, Christopher P. Nelson, Tomasz Czuba, Olle Melander, Margaret Sunitha Selvaraj, Satoshi Koyama, Rohan Bhukar, Yunfeng Ruan, Johan Ljungberg, Scott M. Damrauer, Michael G. Levin, Andre Franke, Klaus Berger, Christian T. Ruff, Giorgio E. M. Melloni, Frederick K. Kamanu, Kaoru Ito, Ron Do, Ruth J. F. Loos, Heribert Schunkert, Quinn S. Wells, Svati H. Shah, Thierry Le Tourneau, David Messika-Zeitoun, Christopher Gignoux, Henning Bundgaard, Susanna C. Larsson, Karl Michaëlsson, Hilma Holm, Anna Helgadottir, Tonu Esko, David A. van Heel, Patrick Mathieu, Nilesh J. Samani, J. Gustav Smith, Stefan Söderberg, Daniel J. Rader, Nicholas A. Marston, Marc S. Sabatine, Bogdan Pasaniuc, Kelly Cho, Peter W. F. Wilson, Christopher J. O’Donnell, Kari Stefansson, Yohan Bossé, Elena Aikawa, James C. Engert, Gina M. Peloso, Pradeep Natarajan, George Thanassoulis <<<

Abstract Aortic stenosis (AS) is a common valvular heart disease and has no pharmacological therapies. We performed a multi-ancestry genome-wide association meta-analysis of 86,864 AS cases among 2,853,408 individuals, discovering 241 autosomal independent risk loci and 3 X chromosome risk loci. We additionally performed sex-stratified and ancestry-stratified genome-wide association studies (GWASs), identifying an additional 5 sex-specific risk loci, 11 risk loci in European ancestry individuals and 1 risk locus in African ancestry individuals. We also performed a transcriptome-wide association study using expression quantitative trait loci from human aortic valves, discovering 54 new genes for which genetically predicted expression influences the risk of AS. We then generated a new polygenic risk score for AS. Finally, we performed gene silencing experiments targeting biologically relevant genes identified by our GWAS. Silencing of CMKLR1 and LTBP4 in human valvular interstitial cells substantially decreased mineralization, implicating a role for polyunsaturated fatty acids and transforming growth factor β signaling in AS. <<<

Vassili Valayannopoulos, Manohar Bance, Daniela S. Carvalho, John H. Greinwald, Steven A. Harvey, Akira Ishiyama, Evie C. Landry, Hubert Löwenheim, Lawrence R. Lustig, Manuel Manrique, Robert Nash, Rubén Polo, Cedric V. Pritchett, Jay T. Rubinstein, A. Eliot Shearer, Ignacio del Castillo, Jeffery J. Anderson, C. Eduardo Corrales, Tera M. Quigley, William J. Riggs, Peter Weber, Gary Wilson, Susan C. Irvin, Hazem E. Hassan, Yanping Chen, Rong Liu, Meghan C. Drummond, Leah R. Sabin, Bret J. Musser, George D. Yancopoulos, Christos A. Kyratsous, Gary A. Herman, Aris Baras, Jonathon P. Whitton <<<

Yuanying Peng, Honghai Yan, Laichun Guo, Cao Deng, Chunlong Wang, Yubo Wang, Lipeng Kang, Pingping Zhou, Kaiquan Yu, Xiaolong Dong, Xiaomeng Liu, Zongyi Sun, Yun Peng, Jun Zhao, Di Deng, Yinghong Xu, Ying Li, Qiantao Jiang, Yan Li, Liming Wei, Jirui Wang, Jian Ma, Ming Hao, Wei Li, Houyang Kang, Zhengsong Peng, Dengcai Liu, Jizeng Jia, Youliang Zheng, Tao Ma, Yuming Wei, Fei Lu, Changzhong Ren <<<

Abstract Common oat ( Avena sativa ) is an important cereal crop serving as a valuable source of forage and human food. Although reference genomes of many important crops have been generated, such work in oat has lagged behind, primarily owing to its large, repeat-rich polyploid genome. Here, using Oxford Nanopore ultralong sequencing and Hi-C technologies, we have generated a reference-quality genome assembly of hulless common oat, comprising 21 pseudomolecules with a total length of 10.76 Gb and contig N50 of 75.27 Mb. We also produced genome assemblies for diploid and tetraploid Avena ancestors, which enabled the identification of oat subgenomes and provided insights into oat chromosomal evolution. The origin of hexaploid oat is inferred from whole-genome sequencing, chloroplast genomes and transcriptome assemblies of different Avena species. These findings and the high-quality reference genomes presented here will facilitate the full use of crop genetic resources to accelerate oat improvement. <<<

Lucas Ferreira DaSilva, Simon Senan, Judith F. Kribelbauer-Swietek, Zain Munir Patel, Lithin Karmel Louis, Aniketh Janardhan Reddy, Sameer Gabbita, Jonathan D. Rosen, Zach Nussbaum, César Miguel Valdez Córdova, Aaron Wenteler, Noah Weber, Tin M. Tunjic, Martino Mansoldo, Talha Ahmad Khan, Gue-Ho Hwang, Vincent Gardeux, David T. Humphreys, Cameron Smith, Matei Bejan, Peter Bromley, Will Connell, Bart Deplancke, Michael I. Love, Emily S. Wong, Wouter Meuleman, Luca Pinello <<<

Charles Findling, Félix Hubert, , Luigi Acerbi, Brandon Benson, Julius Benson, Daniel Birman, Niccolò Bonacchi, E. Kelly Buchanan, Sebastian Bruijns, Matteo Carandini, Joana A. Catarino, Gaelle A. Chapuis, Anne K. Churchland, Yang Dan, Felicia Davatolhagh, Eric E. J. DeWitt, Tatiana A. Engel, Michele Fabbri, Mayo A. Faulkner, Ila Rani Fiete, Laura Freitas-Silva, Berk Gerçek, Kenneth D. Harris, Michael Häusser, Sonja B. Hofer, Fei Hu, Julia M. Huntenburg, Anup Khanal, Chris Krasniak, Christopher Langdon, Christopher A. Langfield, Peter E. Latham, Petrina Y. P. Lau, Zach Mainen, Guido T. Meijer, Nathaniel J. Miska, Thomas D. Mrsic-Flogel, Jean-Paul Noel, Kai Nylund, Alejandro Pan-Vazquez, Liam Paninski, Jonathan Pillow, Cyrille Rossant, Noam Roth, Rylan Schaeffer, Michael Schartner, Yanliang Shi, Karolina Z. Socha, Nicholas A. Steinmetz, Karel Svoboda, Charline Tessereau, Anne E. Urai, Miles J. Wells, Steven Jon West, Matthew R. Whiteway, Olivier Winter, Ilana B. Witten, Anthony Zador, Yizi Zhang, Peter Dayan, Alexandre Pouget <<<

Abstract The neural representations of prior information about the state of the world are poorly understood 1 . Here, to investigate them, we examined brain-wide Neuropixels recordings and widefield calcium imaging collected by the International Brain Laboratory. Mice were trained to indicate the location of a visual grating stimulus, which appeared on the left or right with a prior probability alternating between 0.2 and 0.8 in blocks of variable length. We found that mice estimate this prior probability and thereby improve their decision accuracy. Furthermore, we report that this subjective prior is encoded in at least 20% to 30% of brain regions that, notably, span all levels of processing, from early sensory areas (the lateral geniculate nucleus and primary visual cortex) to motor regions (secondary and primary motor cortex and gigantocellular reticular nucleus) and high-level cortical regions (the dorsal anterior cingulate area and ventrolateral orbitofrontal cortex). This widespread representation of the prior is consistent with a neural model of Bayesian inference involving loops between areas, as opposed to a model in which the prior is incorporated only in decision-making areas. This study offers a brain-wide perspective on prior encoding at cellular resolution, underscoring the importance of using large-scale recordings on a single standardized task. <<<

Few classical games have been regarded as such significant benchmarks of artificial intelligence as to have justified training costs in the millions of dollars. Among these, Stratego -- a board wargame exemplifying the challenge of strategic decision making under massive amounts of hidden information -- stands apart as a case where such efforts failed to produce performance at the level of top humans. This work establishes a step change in both performance and cost for Stratego, showing that it is now possible not only to reach the level of top humans, but to achieve vastly superhuman level -- and that doing so requires not an industrial budget, but merely a few thousand dollars. We achieved this result by developing general approaches for self-play reinforcement learning and test-time search under imperfect information. <<<

Huakang Li, Zixuan Zhao, Meiling Shi, Bin Luo, Guangdong Wang, Xiaojuan Wang, Jie Gu, Zilin Song, Yifan Sun, Li Zhang, Jia Wang <<<

Position encoding recently has shown effective in the transformer architecture. It enables valuable supervision for dependency modeling between elements at different positions of the sequence. In this paper, we first investigate various methods to integrate positional information into the learning process of transformer-based language models. Then, we propose a novel method named Rotary Position Embedding(RoPE) to effectively leverage the positional information. Specifically, the proposed RoPE encodes the absolute position with a rotation matrix and meanwhile incorporates the explicit relative position dependency in self-attention formulation. Notably, RoPE enables valuable properties, including the flexibility of sequence length, decaying inter-token dependency with increasing relative distances, and the capability of equipping the linear self-attention with relative position encoding. Finally, we evaluate the enhanced transformer with rotary position embedding, also called RoFormer, on various long text classification benchmark datasets. Our experiments show that it consistently overcomes its alternatives. Furthermore, we provide a theoretical analysis to explain some experimental results. RoFormer is already integrated into Huggingface: \url{https://huggingface.co/docs/transformers/model_doc/roformer}. <<<