Julien Perolat, Bart De Vylder, Daniel Hennes, Eugene Tarassov, Florian Strub, Vincent de Boer, Paul Muller, Jerome T Connor, Neil Burch, Thomas Anthony, Stephen McAleer, Romuald Elie, Sarah H Cen, Zhe Wang, Audrunas Gruslys, Aleksandra Malysheva, Mina Khan, Sherjil Ozair, Finbarr Timbers, Toby Pohlen, Tom Eccles, Mark Rowland, Marc Lanctot, Jean-Baptiste Lespiau, Bilal Piot, Shayegan Omidshafiei, Edward Lockhart, Laurent Sifre, Nathalie Beauguerlange, Remi Munos, David Silver, Satinder Singh, Demis Hassabis, Karl Tuyls <<<

We introduce DeepNash, an autonomous agent that plays the imperfect information game Stratego at a human expert level. Stratego is one of the few iconic board games that artificial intelligence (AI) has not yet mastered. It is a game characterized by a twin challenge: It requires long-term strategic thinking as in chess, but it also requires dealing with imperfect information as in poker. The technique underpinning DeepNash uses a game-theoretic, model-free deep reinforcement learning method, without search, that learns to master Stratego through self-play from scratch. DeepNash beat existing state-of-the-art AI methods in Stratego and achieved a year-to-date (2022) and all-time top-three ranking on the Gravon games platform, competing with human expert players. <<<

The pervasive uncertainty and dynamic nature of real-world environments present significant challenges for the widespread implementation of machine-driven Intelligent Decision-Making (IDM) systems. Consequently, IDM should possess the ability to continuously acquire new skills and effectively generalize across a broad range of applications. The advancement of Artificial General Intelligence (AGI) that transcends task and application boundaries is critical for enhancing IDM. Recent studies have extensively investigated the Transformer neural architecture as a foundational model for various tasks, including computer vision, natural language processing, and reinforcement learning. We propose that a Foundation Decision Model (FDM) can be developed by formulating diverse decision-making tasks as sequence decoding tasks using the Transformer architecture, offering a promising solution for expanding IDM applications in complex real-world situations. In this paper, we discuss the efficiency and generalization improvements offered by a foundation decision model for IDM and explore its potential applications in multi-agent game AI, production scheduling, and robotics tasks. Lastly, we present a case study demonstrating our FDM implementation, DigitalBrain (DB1) with 1.3 billion parameters, achieving human-level performance in 870 tasks, such as text generation, image captioning, video game playing, robotic control, and traveling salesman problems. As a foundation decision model, DB1 represents an initial step toward more autonomous and efficient real-world IDM applications. <<<

Recent neuroimaging evidence challenges the classical view that face identity and facial expression are processed by segregated neural pathways, showing that information about identity and expression are encoded within common brain regions. This article tests the hypothesis that integrated representations of identity and expression arise spontaneously within deep neural networks. A subset of the CelebA dataset is used to train a deep convolutional neural network (DCNN) to label face identity (chance = 0.06%, accuracy = 26.5%), and the FER2013 dataset is used to train a DCNN to label facial expression (chance = 14.2%, accuracy = 63.5%). The identity-trained and expression-trained networks each successfully transfer to labeling both face identity and facial expression on the Karolinska Directed Emotional Faces dataset. This study demonstrates that DCNNs trained to recognize face identity and DCNNs trained to recognize facial expression spontaneously develop representations of facial expression and face identity, respectively. Furthermore, a congruence coefficient analysis reveals that features distinguishing between identities and features distinguishing between expressions become increasingly orthogonal from layer to layer, suggesting that deep neural networks disentangle representational subspaces corresponding to different sources. <<<

Osteoporosis is a systemic skeletal disease characterized by low bone mass and micro-architectural deterioration of bone tissue, with a consequent increase in bone fragility and fracture susceptibility. In an aged society with increased life expectancy, the incidence rate of osteoporosis is also rapidly increasing. Inadequate nutrition may negatively influence bone metabolism. Recently, many studies have investigated the functionality of milk-derived exosomes, which play important roles in cell-to-cell communication. However, there are few reports of how milk-derived exosomes influence osteoblast proliferation and differentiation. Here, we determined whether bovine colostrum-derived exosomes promote anti-osteoporosis in vitro and in vivo. Tartrate-resistant acid phosphatase-stained cells were significantly inhibited in Raw264.7 cells treated with exosomes, indicating reduced osteoclast differentiation. We induced osteoporosis in mice using glucocorticoid pellets after orally administering exosomes for 2 mo. Interestingly, the bone mineral density of exosome-fed mouse groups was significantly improved compared with the glucocorticoid-induced osteoporosis group without exosome treatment. In addition, Lactobacillus were decreased in the gut microbiota community of osteoporosis-induced mice, but the gut microbiota community composition was effectively restored by exosome intake. Taken together, we propose that exosomes isolated from bovine colostrum could be a potential candidate for osteoporosis prevention, bone remodeling improvement, and inhibition of bone resorption. To our knowledge, this is the first time that a protective effect of milk exosomes against osteoporosis has been demonstrated in vivo. Our results strongly suggest that bovine colostrum exosomes might be used as a prophylaxis to prevent the onset of osteoporosis. Indeed, our results offer promising alternative strategies in the nutritional management of age-related bone complications. <<<

Diffusion probabilistic models have been shown to generate state-of-the-art results on several competitive image synthesis benchmarks but lack a low-dimensional, interpretable latent space, and are slow at generation. On the other hand, standard Variational Autoencoders (VAEs) typically have access to a low-dimensional latent space but exhibit poor sample quality. We present DiffuseVAE, a novel generative framework that integrates VAE within a diffusion model framework, and leverage this to design novel conditional parameterizations for diffusion models. We show that the resulting model equips diffusion models with a low-dimensional VAE inferred latent code which can be used for downstream tasks like controllable synthesis. The proposed method also improves upon the speed vs quality tradeoff exhibited in standard unconditional DDPM/DDIM models (for instance, FID of 16.47 vs 34.36 using a standard DDIM on the CelebA-HQ-128 benchmark using T=10 reverse process steps) without having explicitly trained for such an objective. Furthermore, the proposed model exhibits synthesis quality comparable to state-of-the-art models on standard image synthesis benchmarks like CIFAR-10 and CelebA-64 while outperforming most existing VAE-based methods. Lastly, we show that the proposed method exhibits inherent generalization to different types of noise in the conditioning signal. For reproducibility, our source code is publicly available at this https URL. <<<

Cancer management, including supportive care, is complex and requires availability and synthesis of published and patient-specific data to make appropriate therapeutic decisions. Clinical decision support (CDS) may be an effective implementation strategy to support complex decision making although it is unclear whether it improves process outcomes, patient outcomes or both in cancer settings. We therefore conducted a systematic review to identify CDS that have been used to support therapeutic decision making in clinical cancer settings. Outcomes of interest included the effect of CDS on the process, such as clinician's decision making and effect on patient outcomes. Ten studies met inclusion criteria, with variability in the study design, setting, and intervention. Of the nine studies that measured process outcomes, five demonstrated significant improvement; and of the six that measured patient outcomes, four demonstrated significant improvement. All included studies utilized CDS that were informed by clinical practice guidelines. In conclusion, CDS to guide cancer therapeutic decision making is an understudied but promising area. Further research is needed. <<<

Molecular design strategies are integral to therapeutic progress in drug discovery. Computational approaches for de novo molecular design have been developed over the past three decades and, recently, thanks in part to advances in machine learning (ML) and artificial intelligence (AI), the drug discovery field has gained practical experience. Here, we review these learnings and present de novo approaches according to the coarseness of their molecular representation: that is, whether molecular design is modeled on an atom-based, fragment-based, or reaction-based paradigm. Furthermore, we emphasize the value of strong benchmarks, describe the main challenges to using these methods in practice, and provide a viewpoint on further opportunities for exploration and challenges to be tackled in the upcoming years. <<<

Rina Kim, Ayumi Hashimoto, Nune Markosyan, Vladimir A Tyurin, Yulia Y Tyurina, Gozde Kar, Shuyu Fu, Mohit Sehgal, Laura Garcia-Gerique, Andrew Kossenkov, Bereket A Gebregziabher, John W Tobias, Kristin Hicks, Rebecca A Halpin, Nevena Cvetesic, Hui Deng, Laxminarasimha Donthireddy, Andrew Greenberg, Brian Nam, Robert H Vonderheide, Yulia Nefedova, Valerian E Kagan, Dmitry I Gabrilovich <<<

Ferroptosis is a non-apoptotic form of regulated cell death that is triggered by the discoordination of regulatory redox mechanisms culminating in massive peroxidation of polyunsaturated phospholipids. Ferroptosis inducers have shown considerable effectiveness in killing tumour cells in vitro, yet there has been no obvious success in experimental animal models, with the notable exception of immunodeficient mice. This suggests that the effect of ferroptosis on immune cells remains poorly understood. Pathologically activated neutrophils (PMNs), termed myeloid-derived suppressor cells (PMN-MDSCs), are major negative regulators of anti-tumour immunity. Here we found that PMN-MDSCs in the tumour microenvironment spontaneously die by ferroptosis. Although decreasing the presence of PMN-MDSCs, ferroptosis induces the release of oxygenated lipids and limits the activity of human and mouse T cells. In immunocompetent mice, genetic and pharmacological inhibition of ferroptosis abrogates suppressive activity of PMN-MDSCs, reduces tumour progression and synergizes with immune checkpoint blockade to suppress the tumour growth. By contrast, induction of ferroptosis in immunocompetent mice promotes tumour growth. Thus, ferroptosis is a unique and targetable immunosuppressive mechanism of PMN-MDSCs in the tumour microenvironment that can be pharmacologically modulated to limit tumour progression. <<<

Development of wheat ( L) grain mainly depends on the processes of starch synthesis and storage protein accumulation, which are critical for grain yield and quality. However, the regulatory network underlying the transcriptional and physiological changes of grain development is still not clear. Here, we combined ATAC-seq and RNA-seq to discover the chromatin accessibility and gene expression dynamics during these processes. We found that the chromatin accessibility changes are tightly associated with differential transcriptomic expressions, and the proportion of distal ACRs was increased gradually during grain development. Specific transcription factor (TF) binding sites were enriched at different stages and were diversified among the 3 subgenomes. We further predicted the potential interactions between key TFs and genes related with starch and storage protein biosynthesis and found different copies of some key TFs played diversified roles. Overall, our findings have provided numerous resources and illustrated the regulatory network during wheat grain development, which would shed light on the improvement of wheat yields and qualities. <<<

Chao Xie, Shitong Xiang, Chun Shen, Xuerui Peng, Jujiao Kang, Yuzhu Li, Wei Cheng, Shiqi He, Marina Bobou, M John Broulidakis, Betteke Maria van Noort, Zuo Zhang, Lauren Robinson, Nilakshi Vaidya, Jeanne Winterer, Yuning Zhang, Sinead King, Tobias Banaschewski, Gareth J Barker, Arun L W Bokde, Uli Bromberg, Christian Büchel, Herta Flor, Antoine Grigis, Hugh Garavan, Penny Gowland, Andreas Heinz, Bernd Ittermann, Hervé Lemaître, Jean-Luc Martinot, Marie-Laure Paillère Martinot, Frauke Nees, Dimitri Papadopoulos Orfanos, Tomáš Paus, Luise Poustka, Juliane H Fröhner, Ulrike Schmidt, Julia Sinclair, Michael N Smolka, Argyris Stringaris, Henrik Walter, Robert Whelan, Sylvane Desrivières, Barbara J Sahakian, Trevor W Robbins, Gunter Schumann, Tianye Jia, Jianfeng Feng, IMAGEN Consortium, STRATIFY/ESTRA Consortium, ZIB Consortium <<<

Recent studies proposed a general psychopathology factor underlying common comorbidities among psychiatric disorders. However, its neurobiological mechanisms and generalizability remain elusive. In this study, we used a large longitudinal neuroimaging cohort from adolescence to young adulthood (IMAGEN) to define a neuropsychopathological (NP) factor across externalizing and internalizing symptoms using multitask connectomes. We demonstrate that this NP factor might represent a unified, genetically determined, delayed development of the prefrontal cortex that further leads to poor executive function. We also show this NP factor to be reproducible in multiple developmental periods, from preadolescence to early adulthood, and generalizable to the resting-state connectome and clinical samples (the ADHD-200 Sample and the Stratify Project). In conclusion, we identify a reproducible and general neural basis underlying symptoms of multiple mental health disorders, bridging multidimensional evidence from behavioral, neuroimaging and genetic substrates. These findings may help to develop new therapeutic interventions for psychiatric comorbidities. <<<

SummaryVariable selection plays an important role in high dimensional statistical modelling which nowadays appears in many areas and is key to various scientific discoveries. For problems of large scale or dimensionality p, accuracy of estimation and computational cost are two top concerns. Recently, Candes and Tao have proposed the Dantzig selector using L1-regularization and showed that it achieves the ideal risk up to a logarithmic factor log(p). Their innovative procedure and remarkable result are challenged when the dimensionality is ultrahigh as the factor log(p) can be large and their uniform uncertainty principle can fail. Motivated by these concerns, we introduce the concept of sure screening and propose a sure screening method that is based on correlation learning, called sure independence screening, to reduce dimensionality from high to a moderate scale that is below the sample size. In a fairly general asymptotic framework, correlation learning is shown to have the sure screening property for even exponentially growing dimensionality. As a methodological extension, iterative sure independence screening is also proposed to enhance its finite sample performance. With dimension reduced accurately from high to below sample size, variable selection can be improved on both speed and accuracy, and can then be accomplished by a well-developed method such as smoothly clipped absolute deviation, the Dantzig selector, lasso or adaptive lasso. The connections between these penalized least squares methods are also elucidated. <<<

Jiaqi Hu, Michael Fang, James R Pike, Pamela L Lutsey, A Richey Sharrett, Lynne E Wagenknecht, Timothy M Hughes, Jesse C Seegmiller, Rebecca F Gottesman, Thomas H Mosley, Josef Coresh, Elizabeth Selvin <<<

AIMS/HYPOTHESIS: The aim of this work was to evaluate whether the association of prediabetes with dementia is explained by the intervening onset of diabetes.METHODS: Among participants of the Atherosclerosis Risk in Communities (ARIC) study we defined baseline prediabetes as HbA1c 39-46 mmol/mol (5.7-6.4%) and subsequent incident diabetes as a self-reported physician diagnosis or use of diabetes medication. Incident dementia was ascertained via active surveillance and adjudicated. We quantified the association of prediabetes with dementia risk before and after accounting for the subsequent development of diabetes among ARIC participants without diabetes at baseline (1990-1992; participants aged 46-70 years). We also evaluated whether age at diabetes diagnosis modified the risk of dementia.RESULTS: Among 11,656 participants without diabetes at baseline, 2330 (20.0%) had prediabetes. Before accounting for incident diabetes, prediabetes was significantly associated with the risk of dementia (HR 1.12 [95% CI 1.01, 1.24]). After accounting for incident diabetes, the association was attenuated and non-significant (HR 1.05 [95% CI 0.94, 1.16]). Earlier age of onset of diabetes had the strongest association with dementia: HR 2.92 (95% CI 2.06, 4.14) for onset before 60 years; HR 1.73 (95% CI 1.47, 2.04) for onset at 60-69 years; and HR 1.23 (95% CI 1.08, 1.40) for onset at 70-79 years.CONCLUSIONS/INTERPRETATION: Prediabetes is associated with dementia risk but this risk is explained by the subsequent development of diabetes. Earlier age of onset of diabetes substantially increases dementia risk. Preventing or delaying progression of prediabetes to diabetes will reduce dementia burden. <<<

Dimension reduction techniques for multivariate time series decompose the observed series into a few useful independent/orthogonal univariate components. We develop a spectral domain method for multivariate second-order stationary time series that linearly transforms the observed series into several groups of lower-dimensional multivariate subseries. These multivariate subseries have non-zero spectral coherence among components within a group but have zero spectral coherence among components across groups. The observed series is expressed as a sum of frequency components whose variances are proportional to the spectral matrices at the respective frequencies. The demixing matrix is then estimated using an eigendecomposition on the sum of the variance matrices of these frequency components and its asymptotic properties are derived. Finally, a consistent test on the cross-spectrum of pairs of components is used to find the desired segmentation into the lower-dimensional subseries. The numerical performance of the proposed method is illustrated through simulation examples and an application to modeling and forecasting wind data is presented. <<<

David Ochoa, Andrew Hercules, Miguel Carmona, Daniel Suveges, Asier Gonzalez-Uriarte, Cinzia Malangone, Alfredo Miranda, Luca Fumis, Denise Carvalho-Silva, Michaela Spitzer, Jarrod Baker, Javier Ferrer, Arwa Raies, Olesya Razuvayevskaya, Adam Faulconbridge, Eirini Petsalaki, Prudence Mutowo, Sandra Machlitt-Northen, Gareth Peat, Elaine McAuley, Chuang Kee Ong, Edward Mountjoy, Maya Ghoussaini, Andrea Pierleoni, Eliseo Papa, Miguel Pignatelli, Gautier Koscielny, Mohd Karim, Jeremy Schwartzentruber, David G Hulcoop, Ian Dunham, Ellen M McDonagh <<<

The Open Targets Platform (https://www.targetvalidation.org/) provides users with a queryable knowledgebase and user interface to aid systematic target identification and prioritisation for drug discovery based upon underlying evidence. It is publicly available and the underlying code is open source. Since our last update two years ago, we have had 10 releases to maintain and continuously improve evidence for target-disease relationships from 20 different data sources. In addition, we have integrated new evidence from key datasets, including prioritised targets identified from genome-wide CRISPR knockout screens in 300 cancer models (Project Score), and GWAS/UK BioBank statistical genetic analysis evidence from the Open Targets Genetics Portal. We have evolved our evidence scoring framework to improve target identification. To aid the prioritisation of targets and inform on the potential impact of modulating a given target, we have added evaluation of post-marketing adverse drug reactions and new curated information on target tractability and safety. We have also developed the user interface and backend technologies to improve performance and usability. In this article, we describe the latest enhancements to the Platform, to address the fundamental challenge that developing effective and safe drugs is difficult and expensive. <<<

Maxime Dhainaut, Samuel A Rose, Guray Akturk, Aleksandra Wroblewska, Sebastian R Nielsen, Eun Sook Park, Mark Buckup, Vladimir Roudko, Luisanna Pia, Robert Sweeney, Jessica Le Berichel, C Matthias Wilk, Anela Bektesevic, Brian H Lee, Nina Bhardwaj, Adeeb H Rahman, Alessia Baccarini, Sacha Gnjatic, Dana Pe'er, Miriam Merad, Brian D Brown <<<

While CRISPR screens are helping uncover genes regulating many cell-intrinsic processes, existing approaches are suboptimal for identifying extracellular gene functions, particularly in the tissue context. Here, we developed an approach for spatial functional genomics called Perturb-map. We applied Perturb-map to knock out dozens of genes in parallel in a mouse model of lung cancer and simultaneously assessed how each knockout influenced tumor growth, histopathology, and immune composition. Moreover, we paired Perturb-map and spatial transcriptomics for unbiased analysis of CRISPR-edited tumors. We found that in Tgfbr2 knockout tumors, the tumor microenvironment (TME) was converted to a fibro-mucinous state, and T cells excluded, concomitant with upregulated TGFβ and TGFβ-mediated fibroblast activation, indicating that TGFβ-receptor loss on cancer cells increased TGFβ bioavailability and its immunosuppressive effects on the TME. These studies establish Perturb-map for functional genomics within the tissue at single-cell resolution with spatial architecture preserved and provide insight into how TGFβ responsiveness of cancer cells can affect the TME. <<<

Jessica Roelands, Peter J K Kuppen, Eiman I Ahmed, Raghvendra Mall, Tariq Masoodi, Parul Singh, Gianni Monaco, Christophe Raynaud, Noel F C C de Miranda, Luigi Ferraro, Tatiana C Carneiro-Lobo, Najeeb Syed, Arun Rawat, Amany Awad, Julie Decock, William Mifsud, Lance D Miller, Shimaa Sherif, Mahmoud G Mohamed, Darawan Rinchai, Marc Van den Eynde, Rosalyn W Sayaman, Elad Ziv, Francois Bertucci, Mahir Abdulla Petkar, Stephan Lorenz, Lisa Sara Mathew, Kun Wang, Selvasankar Murugesan, Damien Chaussabel, Alexander L Vahrmeijer, Ena Wang, Anna Ceccarelli, Khalid A Fakhro, Gabriele Zoppoli, Alberto Ballestrero, Rob A E M Tollenaar, Francesco M Marincola, Jérôme Galon, Souhaila Al Khodor, Michele Ceccarelli, Wouter Hendrickx, Davide Bedognetti <<<

The lack of multi-omics cancer datasets with extensive follow-up information hinders the identification of accurate biomarkers of clinical outcome. In this cohort study, we performed comprehensive genomic analyses on fresh-frozen samples from 348 patients affected by primary colon cancer, encompassing RNA, whole-exome, deep T cell receptor and 16S bacterial rRNA gene sequencing on tumor and matched healthy colon tissue, complemented with tumor whole-genome sequencing for further microbiome characterization. A type 1 helper T cell, cytotoxic, gene expression signature, called Immunologic Constant of Rejection, captured the presence of clonally expanded, tumor-enriched T cell clones and outperformed conventional prognostic molecular biomarkers, such as the consensus molecular subtype and the microsatellite instability classifications. Quantification of genetic immunoediting, defined as a lower number of neoantigens than expected, further refined its prognostic value. We identified a microbiome signature, driven by Ruminococcus bromii, associated with a favorable outcome. By combining microbiome signature and Immunologic Constant of Rejection, we developed and validated a composite score (mICRoScore), which identifies a group of patients with excellent survival probability. The publicly available multi-omics dataset provides a resource for better understanding colon cancer biology that could facilitate the discovery of personalized therapeutic approaches. <<<

Asian hair is known for its straightness, dark pigmentation, and large diameter. The cuticle layer in Asians is thicker with more compact cuticle cells than that in Caucasians. Asian hair generally exhibits the strongest mechanical properties, and its cross-sectional area is determined greatly by genetic variations, particularly from the gene. However, knowledge on Asian hair remains unclear with limited studies. This article aimed to review and summarize the characteristics and properties of Asian hair. It also aimed to discuss hair disorders including linear lupus panniculitis and pseudocyst of the scalp that occur distinctively in Asian populations. <<<

Mingkun Lu , Jiayi Yin , Qi Zhu , Gaole Lin , Minjie Mou , Fuyao Liu , Ziqi Pan , Nanxin You , Xichen Lian , Fengcheng Li , Hongning Zhang , Lingyan Zheng , Wei Zhang , Hanyu Zhang , Zihao Shen , Zhen Gu , Honglin Li , Feng Zhu <<<

Drug discovery and development affects various aspects of human health and dramatically impacts the pharmaceutical market. However, investments in a new drug often go unrewarded due to the long and complex process of drug research and development (R&D). With the advancement of experimental technology and computer hardware, artificial intelligence (AI) has recently emerged as a leading tool in analyzing abundant and high-dimensional data. Explosive growth in the size of biomedical data provides advantages in applying AI in all stages of drug R&D. Driven by big data in biomedicine, AI has led to a revolution in drug R&D, due to its ability to discover new drugs more efficiently and at lower cost. This review begins with a brief overview of common AI models in the field of drug discovery; then, it summarizes and discusses in depth their specific applications in various stages of drug R&D, such as target discovery, drug discovery and design, preclinical research, automated drug synthesis, and influences in the pharmaceutical market. Finally, the major limitations of AI in drug R&D are fully discussed and possible solutions are proposed. <<<

Andrew J Stout, Miles J Arnett, Kristin Chai, Tina Guo, Lishu Liao, Addison B Mirliani, Miriam L Rittenberg, Michelle Shub, Eugene C White, John S K Yuen, Xiaoli Zhang, David L Kaplan <<<

For cultured meat to succeed at scale, muscle cells from food-relevant species must be expanded in vitro in a rapid and reliable manner to produce millions of metric tons of biomass annually. Toward this goal, genetically immortalized cells offer substantial benefits over primary cells, including rapid growth, escape from cellular senescence, and consistent starting cell populations for production. Here, we develop genetically immortalized bovine satellite cells (iBSCs) via constitutive expression of bovine Telomerase reverse transcriptase (TERT) and Cyclin-dependent kinase 4 (CDK4). These cells achieve over 120 doublings at the time of publication and maintain their capacity for myogenic differentiation. They therefore offer a valuable tool to the field, enabling further research and development to advance cultured meat. <<<

The regularities of the world render an intricate interplay between past and present. Even across independent trials, current-trial perception can be automatically shifted by preceding trials, namely the "serial bias." Meanwhile, the neural implementation of the spontaneous shift of present by past that operates on multiple features remains unknown. In two auditory categorization experiments with human electrophysiological recordings, we demonstrate that serial bias arises from the co-occurrence of past-trial neural reactivation and the neural encoding of current-trial features. The meeting of past and present shifts the neural representation of current-trial features and modulates serial bias behavior. Critically, past-trial features (i.e., pitch, category choice, motor response) keep their respective identities in memory and are only reactivated by the corresponding features in the current trial, giving rise to dissociated feature-specific serial biases. The feature-specific automatic reactivation might constitute a fundamental mechanism for adaptive past-to-present generalizations over multiple features. <<<