David W. Romero

Abstract All of life encodes information with DNA. While tools for sequencing, synthesis, and editing of genomic code have transformed biological research, intelligently composing new biological systems would also require a deep understanding of the immense complexity encoded by genomes. We introduce Evo 2, a biological foundation model trained on 9.3 trillion DNA base pairs from a highly curated genomic atlas spanning all domains of life. We train Evo 2 with 7B and 40B parameters to have an unprecedented 1 million token context window with single-nucleotide resolution. Evo 2 learns from DNA sequence alone to accurately predict the functional impacts of genetic variation—from noncoding pathogenic mutations to clinically significant BRCA1 variants—without task-specific finetuning. Applying mechanistic interpretability analyses, we reveal that Evo 2 autonomously learns a breadth of biological features, including exon–intron boundaries, transcription factor binding sites, protein structural elements, and prophage genomic regions. Beyond its predictive capabilities, Evo 2 generates mitochondrial, prokaryotic, and eukaryotic sequences at genome scale with greater naturalness and coherence than previous methods. Guiding Evo 2 via inference-time search enables controllable generation of epigenomic structure, for which we demonstrate the first inference-time scaling results in biology. We make Evo 2 fully open, including model parameters, training code, inference code, and the OpenGenome2 dataset, to accelerate the exploration and design of biological complexity.

Most applications of spatial modeling to the problem of electoral competition are pessimistic regarding the prospects for candidate equilibrium in more than one policy dimension. Probabilistic models of the vote, however, increase the likelihood of equilibrium. We expand the probabilistic model to include measured nonissue variables, thereby representing the general multivariate model of behavioral research. For this model we offer a general candidate equilibrium solution and illustrate with some simulations based on 1988 National Election Study data. The more complicated one's model of voters' motivations, the greater appears to be the chance of locating a candidate equilibrium position in policy space.

Except for bivariate analyses, previous research on the group basis of partisan strength in the United States has focused on party identification as the dependent variable. This essay examines the group basis of the presidential vote, 1952-1984, using a multivariate logit approach. Our multivariate analysis shows the persistence of group-based divisions between Republican and Democratic voters. Among other patterns, class-based divisions have noticeably increased.

Although group convolutional networks are able to learn powerful representations based on symmetry patterns, they lack explicit means to learn meaningful relationships among them (e.g., relative positions and poses). In this paper, we present attentive group equivariant convolutions, a generalization of the group convolution, in which attention is applied during the course of convolution to accentuate meaningful symmetry combinations and suppress non-plausible, misleading ones. We indicate that prior work on visual attention can be described as special cases of our proposed framework and show empirically that our attentive group equivariant convolutional networks consistently outperform conventional group convolutional networks on benchmark image datasets. Simultaneously, we provide interpretability to the learned concepts through the visualization of equivariant attention maps.

For nearly two decades congressional elections scholars have struggled to match individual-level findings for the pocketbook voting thesis with ag gregate-level findings. Whereas strong and consistent support can be found for the pocketbook thesis in presidential elections with both individual- level and aggregate-level data, significant support has been found for this thesis in U.S. House elections with aggregate-level data only. The typical individual-level congressional elections study examines a single election or series of single elections. We take an alternative approach. By pooling elections from 1980 to 1990, this inquiry demonstrates that retrospective pocketbook assessments indeed influence the individual's vote in congres sional elections. The benefit this result brings to the elections literature is symmetry between the findings for presidential elections and congressional elections regarding the decision rule voters follow when weighing eco nomic conditions when they vote.