Proceedings of the National Academy of Sciences of the United States of America最新文献

The Golden Horde, the northwestern extension of the Mongol Empire ruled by Genghis Khan's descendants, holds a pivotal place in the history of Central Eurasia and Eastern Europe. Consequently, understanding the genetic legacy of Genghis Khan and his lineage has long been of both academic and public interest, especially concerning the hypothesized association of his Y-chromosome with haplogroup C3*. Here, we present ancient DNA data from four archaeological individuals-three males and one female-from medieval elite mausoleums of the Golden Horde in the Ulitau region of Kazakstan. Our genomic analyses reveal that the three male individuals are paternally related and share the Y-chromosome haplogroup C3*, confirming the association between the Y-chromosome haplogroup C3* and the Mongol Empire, supporting the long-standing hypothesis about the genetic legacy of Mongols. Additionally, our findings demonstrate that the Golden Horde elites primarily derive their genomes from Ancient Northeast Asians (ANA), with an additional ancestral component from either Ancient North Eurasians (ANE) or a Berel Scythian related population, e.g., the Kipchaks. Archaeological evidence, in turn, sheds light on a medieval population undergoing religious and cultural transition, offering insights into the societal changes experienced by Mongolian conquerors. Furthermore, through constructing an Identity by Descent (IBD) network, we successfully identify medieval relatives of these individuals on the Mongolian Plateau, linking genetic data to broader population dynamics. In essence, this study provides ancient DNA evidence that advances our understanding of the genetic background of the Mongolian elites and the population dynamics in Central Eurasia.

The current cosmological paradigm, ΛCDM, is characterized by its expansive description of the history of the Universe, its deep connections to particle physics and the large quantities of data that support it. Nonetheless, ΛCDM's critics argue that it has been falsified or must be discarded for various reasons. Critics and boosters alike do agree on one thing: It is not the final cosmological theory and they are anxious to see it replaced by something better! I review the status of ΛCDM, provide my views on what "better" might look like, and discuss the role that the "Hubble tension" might play in moving beyond ΛCDM.

Self-incompatibility (SI) describes a widespread collection of genetic mechanisms in flowering plants used to specifically recognize and reject self-pollen. These mechanisms are fundamental to plant sexual reproduction and offer valuable insight into the molecular basis of cell-cell communication and self-recognition more broadly. Here, we leverage an independent evolution of SI in the lineage containing Phlox (Polemoniaceae) to identify the gene causing self-pollen recognition which we name Phlox drummondii Pistil Identity Receptor Kinase (PdPIRK). Recognition of self-pollen associates with a single genomic region containing the Phlox S-locus. We generate predictions regarding how S-loci must function and evolve to identify a single candidate gene within this S-associated region. This gene, PdPIRK, is highly and specifically expressed in the pistil and has exceptionally high polymorphism maintained by negative frequency-dependent selection, two hallmarks of self-pollen recognition genes. Functional validation with gene silencing confirms that PdPIRK is necessary for self-incompatibility, and we further demonstrate allele-specific activity, confirming its role in self-pollen recognition per se. PdPIRK encodes a G-type lectin receptor-like kinase, which is a member of the same gene family as SRK, the gene controlling self-pollen recognition in the distantly related Brassicaceae. Our findings suggest the presence of genetic constraints or paths of least resistance governing how S-loci evolve and add to our understanding of the diverse molecular mechanisms through which organisms achieve self-recognition.

Hallucinated facts in large language models have recently been shown to obey a statistical lower bound determined by the monofact rate (related to the classical Good-Turing missing mass estimator) minus model miscalibration [A. T. Kalai, S. S. Vempala, "Calibrated language models must hallucinate" in Proceedings of the 56th Annual ACM Symposium on Theory of Computing (STOC) (New York, NY, USA, 2024), pp. 160-171]. We present empirical investigation of this three-way relationship in classical [Formula: see text]-gram models and fine-tuned transformer models. By generating training data from Pareto distributions with varying shape parameters, we systematically control the monofact rate and establish its positive relationship with hallucination. To bridge theory and practice, we derive an empirical analog of the hallucination bound by replacing the population miscalibration term (Section 1.1) with an empirical bin-wise Kullback-Leibler (KL) divergence and confirm its practical viability. We then introduce selective upweighting-a simple yet effective technique that strategically repeats as little as 5% of training examples-to deliberately inject miscalibration into the model. This intervention reduces hallucination by up to 40%, challenging universal deduplication policies. Our experiments reveal a critical trade-off: selective upweighting maintains preinjection levels of accuracy while substantially reducing hallucination, whereas standard training gradually improves accuracy but fails to address persistently high hallucination, indicating an inherent tension in optimization objectives.