Steven Weaver

Autonomous artificial intelligence agents are transforming the practice of science. Across every discipline, researchers are handing data analysis to coding agents that can design and execute complex studies from a single prompt. Yet this revolution carries a threat. When agents improvise in local environments, they leave no verifiable record, ultimately risking a wave of irreproducible science. Solving this crisis requires moving agentic science onto public infrastructure where every action leaves an examinable trail. Here we connect autonomous coding agents to shared computing resources, turning ephemeral prompts into auditable scientific records. After showing that 30 unguided local agents splinter into 18 incompatible pipelines on the same dataset, we conduct an agentic reproduction and discovery study on a large scale—reproducing a large, diverse array of studies with every workflow, history, and intermediate dataset publicly accessible. This complete provenance trail exposes unstated modeling choices, reference mismatches, and subsampled tutorial inputs in published studies, and establishes a new foundation where human scientists can understand, steer, and trust AI-driven discovery.

Standard probabilistic models of coding sequence evolution effectively identify where and when selection acts but remain agnostic to the mechanistic realization of these forces. We introduce PRIME (PRoperty Informed Models of Evolution), a framework of codon-level maximum likelihood methods—including global (G-PRIME), episodic (E-PRIME), and site-specific (S-PRIME) implementations—that explicitly model amino acid exchangeability as a function of physicochemical properties. By parameterizing attributes such as molecular volume, hydropathy, and secondary structure propensities, PRIME aims to resolve the biophysical basis of selective constraint across both the sequence and the phylogeny. At the site level, S-PRIME leverages an explicit biophysical taxonomy to categorize residues as conserved, neutral, or changing for specific properties, resolving selective signals that are missed by traditional rate-based metrics. Our analysis of a benchmark of 24 diverse datasets and a genome-wide screen of 18,944 mammalian genes demonstrates that consideration of biophysical realism can yield substantial improvements in model fit, acting synergistically with rate variation to explain complex evolutionary patterns. We find that physicochemical constraints at individual sites can be reliably detected in datasets with sufficient information redundancy (substitutions per unique amino acid; AUC=0.91), with sensitivity exceeding 90% in data-rich alignments. E-PRIME reveals a distinct hierarchy in biophysical constraints: while core packing and beta-sheet scaffolds are rigidly conserved, alpha-helix propensity and surface electrostatics serve as the primary substrates for adaptive tuning. Furthermore, PRIME importance weights align with aspects of the primary semantic axes of deep learning representations (ESM-2) and capture key features of experimental fitness landscapes. By transforming abstract evolutionary rates into interpretable biophysical rules, PRIME provides a useful framework for characterizing the mechanistic drivers of protein diversity.

The quantification of genomic conservation has progressed from foundational statistical modeling of evolutionary rates to state-of-the-art deep learning architectures. However, a major resolution gap remains at the zero-rate origin, where standard selection inference tools fail to distinguish between sites that are invariant due to chance (stochastic invariance) or low substitution opportunity and those that are invariant due to extreme purifying selection. We present B-STILL (Bayesian Significance Test of Invariant Low Likelihoods), a hierarchical Bayesian framework designed to resolve the selective landscape of protein-coding genes near the zero-rate limit. By leveraging gene-level rate distributions (prior calibration) and modeling codon-site-specific substitution opportunities (determined by genetic-code degeneracy and nucleotide substitution biases), B-STILL quantifies the statistical significance of observed stasis. We define a rate-based stasis threshold to identify evolutionary stasis anchors (ESAs)—sites where the upper bound on the evolutionary rate is statistically constrained relative to the background rate of the gene due to extreme purifying selection. Validation against clinical and pathogen datasets confirms that ESAs are strong predictors of biological fitness and pathogenicity. Applying B-STILL across viral and mammalian genomes, we identify thousands of significantly clustered ESAs that map to known functional domains and uncharacterized structural motifs. These results establish B-STILL as a scalable, statistically rigorous framework for high-resolution genomic annotation, converting previously uninformative invariant sites into precise markers of extreme evolutionary constraint.

Understanding HIV transmission in densely populated urban settings is essential to mitigate ongoing epidemic spread. We present a comprehensive analysis of recent HIV transmission dynamics in Greater Mexico City, one of the world’s largest metropolitan areas comprising Mexico City and neighbouring municipalities of the State of Mexico. Drawing from over 7,000 complete pol gene sequences representing around 50% of new cases reported between 2019 and 2022 within the study region, we reconstructed the transmission network based on pairwise genetic distance. We identified ten large transmission clusters exhibiting sustained growth up to the most recent sampling period. We further analysed paired genetic and high-resolution human mobility data using an integrated phylogeographic approach. We observed a heterogeneous pattern of viral spread across the region, supported by an extensive mixing at a wider geographic scale. Across Greater Mexico City, displaying a high population density, HIV transmission is minimally spatially constrained, a pattern likely fuelled by intense human mobility. Thus, population movement weakens isolation by distance in large urban areas even for a chronic infection that is sexually and vertically transmitted. We demonstrate the value of integrating large-scale genetic, epidemiological, and mobility data to resolve contemporary HIV transmission dynamics in densely populated urban settings.

Positive selection is an evolutionary process which increases the frequency of advantageous mutations because they confer a fitness benefit. Inferring the past action of positive selection on protein-coding sequences is fundamental for deciphering phenotypic diversity and the emergence of novel traits. With the advent of genome-wide comparative genomic datasets, researchers can analyze selection not only at the level of individual genes but also globally, delivering systems-level insights into evolutionary dynamics. However, genome-scale datasets are generated with automated pipelines and imperfect curation that does not eliminate all sequencing, annotation, and alignment errors. Positive selection inference methods are highly sensitive to such errors. We present BUSTED-E: a method designed to detect positive selection for amino acid diversification while concurrently identifying some alignment errors. This method builds on the flexible branch-site random effects model (BUSTED) for fitting distributions of dN/dS, with a critical modification: it incorporates an “error-sink” component to represent an abiological evolutionary regime. Using several genome-scale biological datasets that were extensively filtered using state-of-the art automated alignment tools, we show that BUSTED-E identifies pervasive residual alignment errors, produces more realistic estimates of positive selection, reduces bias, and improves biological interpretation. The BUSTED-E model promises to be a more stringent filter to identify positive selection in genome-wide contexts, thus enabling further characterization and validation of the most biologically relevant cases.

Positive selection is an evolutionary process which increases the frequency of advantageous mutations because they confer a fitness benefit. Inferring the past action of positive selection on protein-coding sequences is fundamental for deciphering phenotypic diversity and the emergence of novel traits. With the advent of genome-wide comparative genomic datasets, researchers can analyze selection not only at the level of individual genes but also globally, delivering systems-level insights into evolutionary dynamics. However, genome-scale datasets are generated with automated pipelines and imperfect curation that does not eliminate all sequencing, annotation, and alignment errors. Positive selection inference methods are highly sensitive to such errors. We present BUSTED-E: a method designed to detect positive selection for amino acid diversification while concurrently identifying some alignment errors. This method builds on the flexible branch-site random effects model (BUSTED) for fitting distributions of dN/dS, with a critical modification: it incorporates an “error-sink” component to represent an abiological evolutionary regime. Using several genome-scale biological datasets that were extensively filtered using state-of-the art automated alignment tools, we show that BUSTED-E identifies pervasive residual alignment errors, produces more realistic estimates of positive selection, reduces bias, and improves biological interpretation. The BUSTED-E model promises to be a more stringent filter to identify positive selection in genome-wide contexts, thus enabling further characterization and validation of the most biologically relevant cases.

Molecular surveillance of viral pathogens and inference of transmission networks from genomic data play an increasingly important role in public health efforts, especially for HIV-1. For many methods, the genetic distance threshold used to connect sequences in the transmission network is a key parameter informing the properties of inferred networks. Using a distance threshold that is too high can result in a network with many spurious links, making it difficult to interpret. Conversely, a distance threshold that is too low can result in a network with too few links, which may not capture key insights into clusters of public health concern. Published research using the HIV-TRACE software package frequently uses the default threshold of 0.015 substitutions/site for HIV pol gene sequences, but in many cases, investigators heuristically select other threshold parameters to better capture the underlying dynamics of the epidemic they are studying. Here, we present a general heuristic scoring approach for tuning a distance threshold adaptively, which seeks to prevent the formation of giant clusters. We prioritize the ratio of the sizes of the largest and the second largest cluster, maximizing the number of clusters present in the network. We apply our scoring heuristic to outbreaks with different characteristics, such as regional or temporal variability, and demonstrate the utility of using the scoring mechanism’s suggested distance threshold to identify clusters exhibiting risk factors that would have otherwise been more difficult to identify. For example, while we found that a 0.015 substitutions/site distance threshold is typical for US-like epidemics, recent outbreaks like the CRF07_BC subtype among men who have sex with men (MSM) in China have been found to have a lower optimal threshold of 0.005 to better capture the transition from injected drug use (IDU) to MSM as the primary risk factor. Alternatively, in communities surrounding Lake Victoria in Uganda, where there has been sustained hetero-sexual transmission for many years, we found that a larger distance threshold is necessary to capture a more risk factor-diverse population with sparse sampling over a longer period of time. Such identification may allow for more informed intervention action by respective public health officials.

Building reliable phylogenies from very large collections of sequences with a limited number of phylogenetically informative sites is challenging because sequencing errors and recurrent/backward mutations interfere with the phylogenetic signal, confounding true evolutionary relationships. Massive global efforts of sequencing genomes and reconstructing the phylogeny of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) strains exemplify these difficulties since there are only hundreds of phylogenetically informative sites but millions of genomes. For such datasets, we set out to develop a method for building the phylogenetic tree of genomic haplotypes consisting of positions harboring common variants to improve the signal-to-noise ratio for more accurate and fast phylogenetic inference of resolvable phylogenetic features. We present the TopHap approach that determines spatiotemporally common haplotypes of common variants and builds their phylogeny at a fraction of the computational time of traditional methods. We develop a bootstrap strategy that resamples genomes spatiotemporally to assess topological robustness. The application of TopHap to build a phylogeny of 68 057 SARS-CoV-2 genomes (68KG) from the first year of the pandemic produced an evolutionary tree of major SARS-CoV-2 haplotypes. This phylogeny is concordant with the mutation tree inferred using the co-occurrence pattern of mutations and recovers key phylogenetic relationships from more traditional analyses. We also evaluated alternative roots of the SARS-CoV-2 phylogeny and found that the earliest sampled genomes in 2019 likely evolved by four mutations of the most recent common ancestor of all SARS-CoV-2 genomes. An application of TopHap to more than 1 million SARS-CoV-2 genomes reconstructed the most comprehensive evolutionary relationships of major variants, which confirmed the 68KG phylogeny and provided evolutionary origins of major and recent variants of concern. TopHap is available at https://github.com/SayakaMiura/TopHap. Supplementary data are available at Bioinformatics online.

SARS-CoV-2 evolution threatens vaccine- and natural infection–derived immunity and the efficacy of therapeutic antibodies. To improve public health preparedness, we sought to predict which existing amino acid mutations in SARS-CoV-2 might contribute to future variants of concern. We tested the predictive value of features comprising epidemiology, evolution, immunology, and neural network–based protein sequence modeling and identified primary biological drivers of SARS-CoV-2 intrapandemic evolution. We found evidence that ACE2-mediated transmissibility and resistance to population-level host immunity has waxed and waned as a primary driver of SARS-CoV-2 evolution over time. We retroactively identified with high accuracy (area under the receiver operator characteristic curve = 0.92 to 0.97) mutations that will spread, at up to 4 months in advance, across different phases of the pandemic. The behavior of the model was consistent with a plausible causal structure where epidemiological covariates combine the effects of diverse and shifting drivers of viral fitness. We applied our model to forecast mutations that will spread in the future and characterize how these mutations affect the binding of therapeutic antibodies. These findings demonstrate that it is possible to forecast the driver mutations that could appear in emerging SARS-CoV-2 variants of concern. We validated this result against Omicron, showing elevated predictive scores for its component mutations before emergence and rapid score increase across daily forecasts during emergence. This modeling approach may be applied to any rapidly evolving pathogens with sufficiently dense genomic surveillance data, such as influenza, and unknown future pandemic viruses.