RESUMEN
Chip floorplanning is the engineering task of designing the physical layout of a computer chip. Despite five decades of research1, chip floorplanning has defied automation, requiring months of intense effort by physical design engineers to produce manufacturable layouts. Here we present a deep reinforcement learning approach to chip floorplanning. In under six hours, our method automatically generates chip floorplans that are superior or comparable to those produced by humans in all key metrics, including power consumption, performance and chip area. To achieve this, we pose chip floorplanning as a reinforcement learning problem, and develop an edge-based graph convolutional neural network architecture capable of learning rich and transferable representations of the chip. As a result, our method utilizes past experience to become better and faster at solving new instances of the problem, allowing chip design to be performed by artificial agents with more experience than any human designer. Our method was used to design the next generation of Google's artificial intelligence (AI) accelerators, and has the potential to save thousands of hours of human effort for each new generation. Finally, we believe that more powerful AI-designed hardware will fuel advances in AI, creating a symbiotic relationship between the two fields.
RESUMEN
Limited ancestral diversity has impaired our ability to detect risk variants more prevalent in non-European ancestry groups in genome-wide association studies (GWAS). We constructed and analyzed a multi-ancestry GWAS dataset in the Alzheimer's Disease (AD) Genetics Consortium (ADGC) to test for novel shared and ancestry-specific AD susceptibility loci and evaluate underlying genetic architecture in 37,382 non-Hispanic White (NHW), 6,728 African American, 8,899 Hispanic (HIS), and 3,232 East Asian individuals, performing within-ancestry fixed-effects meta-analysis followed by a cross-ancestry random-effects meta-analysis. We identified 13 loci with cross-ancestry associations including known loci at/near CR1 , BIN1 , TREM2 , CD2AP , PTK2B , CLU , SHARPIN , MS4A6A , PICALM , ABCA7 , APOE and two novel loci not previously reported at 11p12 ( LRRC4C ) and 12q24.13 ( LHX5-AS1 ). Reflecting the power of diverse ancestry in GWAS, we observed the SHARPIN locus using 7.1% the sample size of the original discovering single-ancestry GWAS (n=788,989). We additionally identified three GWS ancestry-specific loci at/near ( PTPRK ( P =2.4×10 -8 ) and GRB14 ( P =1.7×10 -8 ) in HIS), and KIAA0825 ( P =2.9×10 -8 in NHW). Pathway analysis implicated multiple amyloid regulation pathways (strongest with P adjusted =1.6×10 -4 ) and the classical complement pathway ( P adjusted =1.3×10 -3 ). Genes at/near our novel loci have known roles in neuronal development ( LRRC4C, LHX5-AS1 , and PTPRK ) and insulin receptor activity regulation ( GRB14 ). These findings provide compelling support for using traditionally-underrepresented populations for gene discovery, even with smaller sample sizes.
RESUMEN
Acute lymphoblastic leukemia (ALL) is the most common childhood cancer. Here, using whole-genome, exome and transcriptome sequencing of 2,754 childhood patients with ALL, we find that, despite a generally low mutation burden, ALL cases harbor a median of four putative somatic driver alterations per sample, with 376 putative driver genes identified varying in prevalence across ALL subtypes. Most samples harbor at least one rare gene alteration, including 70 putative cancer driver genes associated with ubiquitination, SUMOylation, noncoding transcripts and other functions. In hyperdiploid B-ALL, chromosomal gains are acquired early and synchronously before ultraviolet-induced mutation. By contrast, ultraviolet-induced mutations precede chromosomal gains in B-ALL cases with intrachromosomal amplification of chromosome 21. We also demonstrate the prognostic significance of genetic alterations within subtypes. Intriguingly, DUX4- and KMT2A-rearranged subtypes separate into CEBPA/FLT3- or NFATC4-expressing subgroups with potential clinical implications. Together, these results deepen understanding of the ALL genomic landscape and associated outcomes.
Asunto(s)
Leucemia-Linfoma Linfoblástico de Células Precursoras , Niño , Aberraciones Cromosómicas , Exoma/genética , Genómica , Humanos , Mutación , Leucemia-Linfoma Linfoblástico de Células Precursoras/genéticaRESUMEN
Chemotherapy is a standard treatment for pediatric acute lymphoblastic leukemia (ALL), which sometimes relapses with chemoresistant features. However, whether acquired drug-resistance mutations in relapsed ALL pre-exist or are induced by treatment remains unknown. Here we provide direct evidence of a specific mechanism by which chemotherapy induces drug-resistance-associated mutations leading to relapse. Using genomic and functional analysis of relapsed ALL we show that thiopurine treatment in mismatch repair (MMR)-deficient leukemias induces hotspot TP53 R248Q mutations through a specific mutational signature (thio-dMMR). Clonal evolution analysis reveals sequential MMR inactivation followed by TP53 mutation in some patients with ALL. Acquired TP53 R248Q mutations are associated with on-treatment relapse, poor treatment response and resistance to multiple chemotherapeutic agents, which could be reversed by pharmacological p53 reactivation. Our findings indicate that TP53 R248Q in relapsed ALL originates through synergistic mutagenesis from thiopurine treatment and MMR deficiency and suggest strategies to prevent or treat TP53-mutant relapse.