Berkeley's genome model buys accuracy with alignments instead of parameters
GPN-Star folds whole-genome alignments and a species tree into its architecture, which is how it trains in hours where Evo 2 needed 2,000 processors and months. The cost of building those alignments sits outside that comparison.
The Scientist · Science desk
What happened
- UC Berkeley's GPN-Star, published in Nature, is a genomic language model whose architecture takes whole-genome alignments and species trees as explicit inputs rather than learning from unaligned genomes.
- Trained on alignments spanning vertebrate, mammal and primate timescales, it reports state-of-the-art variant effect prediction across both coding and non-coding regions of the human genome.
- The paper also reports substantial gains over previous methods in prioritizing pathogenic and fine-mapped GWAS variants, in complex trait heritability enrichment and in rare variant association power.
- Training took days or even hours on a handful of processors, against the 2,000 NVIDIA processors and months of runtime that the larger Evo 2 model required.
- Versions were trained for five model organisms as well: mouse, chicken, Drosophila melanogaster, Caenorhabditis elegans and Arabidopsis thaliana.
Why it matters
- What it costs cost The cheap training run sits on top of an expensive input: high-quality whole-genome alignments were expanded by large-scale consortia, so the compute saving is partly a transfer of cost to whoever builds and maintains the alignments.
- What it limits constraint Because the model's signal is borrowed from comparative genomics data, its usefulness travels only as far as good alignments already do, which is a real limit for any organism that lacks them.
- Who has to decide decision A group choosing a tool is not weighing equivalents: Evo 2 can generate whole genomes, while GPN-Star scores existing variants, and only the scoring job gets cheaper here.
- What's now possible capability With genome-wide predictions released, a lab that can afford ten assays instead of ten thousand now has a published ranking to spend them against.
Next story · Science · 2d ago
Microplastics generate reactive oxygen species in water, a Nature study from Columbia and Berkeley reportsConfirmed6 publishers
The denominator first: of the human genome's 3 billion base pairs, 1% to 2% codes for proteins [17], which leaves roughly 2.94 to 2.97 billion bases of evolutionary holdover and regulatory sequence [21]. Working out which of those bases matter is what a conservation signal is for. A whole-genome alignment relates hundreds of species to one anchor genome and marks, site by site, where sequence has been conserved and where it has changed [5]. A model trained on unaligned genomes has to recover that correspondence itself; GPN-Star receives it as input, and that is where the training savings come from [10].
The Nature paper is direct about the problem this addresses. Genomic language models built on standard language-modelling frameworks still lose to much simpler classical phylogenetic models on some variant interpretation tasks even at massive model sizes, and the shortfall is worst in complex eukaryotic genomes and in distal regulatory elements such as enhancers [11]. PhastCons and PhyloP, the parametric models fitted on alignments, have been staples of human variant interpretation for years [12]. Proteins met this argument first: alignment-based models such as AlphaFold, MSA Transformer and EVE worked, and interest in alignments revived once scaling single-sequence protein language models showed diminishing returns [13].
Evo 2 ingested more than 100,000 species using 2,000 NVIDIA processors over months [8]. The alignments GPN-Star learns from span hundreds of species [5], so even reading "hundreds" at its upper bound, the alignment route looks at fewer genomes by a factor of at least a hundred [22]. The efficiency claim is about how much raw sequence ever enters the model, not only about parameter count.
Neither the human-genome benchmark result nor the prioritization gain comes with a number in the abstract; the wording is qualitative [23]. Yun Song, the study's senior author [19], places the model upstream of the bench, saying that nobody can experimentally test every variant in the genome and that the predictions should help prioritize the experiments with the greatest potential impact on human health [16]. For a conservation-derived score, that is the defensible position: it ranks candidates, and something else confirms them.
One result deserves more attention than the ranking. Modelling recent evolution and modelling deep evolution help different tasks, and the paper reports that advantage as task-dependent [6]. The vertebrate, mammal and primate versions are therefore three different instruments rather than three rungs of a ladder, and choosing among them falls to whoever runs the model on a particular question.
What to watch
- Independent benchmarks that put a number on GPN-Star's margin over PhyloP, PhastCons and Evo 2, especially in enhancers.
- Experimental follow-up on the published genome-wide predictions: which prioritized variants survive assay.
- Whether the framework transfers to species without high-quality whole-genome alignments already built.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence60
- Adoption14
- Hype gap+28
- Incentives72
- Confidence58
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
Researchers at UC Berkeley created a genomic language model called GPN-Star; the study was published in the journal Nature.
- [2]
GPN-Star (genomic pretrained network with species tree and alignment representations) is a genomic language model with a phylogeny-aware architecture that leverages whole-genome alignments and species trees to model evolutionary relationships explicitly.
- [3]
Trained on alignments spanning vertebrate, mammal and primate evolutionary timescales, GPN-Star achieves state-of-the-art performance across a wide range of variant effect prediction tasks in both coding and non-coding regions of the human genome.
- [4]
The paper reports that GPN-Star substantially outperforms previous methods in prioritizing pathogenic and fine-mapped genome-wide association study variants, yields strong enrichments of complex trait heritability, and improves power in rare variant association testing.
- [5]
Whole-genome alignments use specialized algorithms to relate the genomes of hundreds of different species to that of a single species; in a human-anchored alignment, other genomes are compared to the human genome, revealing where code has been conserved over evolution and where it has changed.
- [6]
Analyses across evolutionary timescales show task-dependent advantages of modelling more recent versus deeper evolution.
- [7]
The authors also trained GPN-Star for five model organisms: Mus musculus, Gallus gallus, Drosophila melanogaster, Caenorhabditis elegans and Arabidopsis thaliana.
- [8]
The Evo 2 model was trained on the genomes of more than 100,000 species across all domains of life and required 2,000 powerful NVIDIA computer processors and months to train.
- [9]
GPN-Star can be trained in just days, or even hours, using only a handful of processors.
- [10]
Because whole-genome alignments do the work of identifying conserved areas of code, GPN-Star takes much less time and computing power to train than models that use unaligned genomes.
- [11]
Genomic language models based on standard language modelling frameworks still underperform much simpler classical phylogenetic models on certain variant interpretation tasks, particularly in complex eukaryotic genomes such as in humans and in distal regulatory elements such as enhancers, even with massive model sizes.
- [12]
Classical parametric phylogenetic models fitted on whole-genome alignments, such as PhastCons and PhyloP, have long been staples in variant interpretation for humans and other species.
- [13]
Protein multiple sequence alignments enabled models including AlphaFold, MSA Transformer and EVE, and have recently experienced renewed interest as scaling single-sequence protein language models has shown diminishing returns.
- [14]
Recent large-scale consortia have expanded high-quality whole-genome alignment resources.
- [15]
Alongside the study, the researchers published genome-wide predictions from the model highlighting genetic variants likely to have the most influence on inherited traits, which biologists can use to identify relevant genes and regulatory elements for further study.
- [16]
Song said that people have developed creative tools for assaying the impact of genetic variants but cannot experimentally test every single variant in the genome, and that the team believes its predictions will help prioritize the experiments that could have the greatest impact on human health.
- [17]
The human genome comprises 3 billion base pairs, of which an estimated 1% to 2% codes for proteins; the rest is a mix of junk DNA and regulatory elements that control when, where and how strongly genes are expressed.
- [18]
Evo 2 can generate entire genomes from scratch.
- [19]
Yun Song, the study's senior author, is a professor of computer science and statistics at UC Berkeley and an investigator at the Innovative Genomics Institute.
- [20]
The paper positions GPN-Star as well suited to leverage the growing abundance of comparative genomics data.
- [21]
Roughly 2.94 billion to 2.97 billion base pairs of the human genome do not code for proteins.
- [22]
GPN-Star's alignment-based training data covers at least a hundredfold fewer genomes than Evo 2's training set.
- [23]
The Nature abstract states GPN-Star's performance qualitatively, as 'state-of-the-art' and as substantially outperforming previous methods, without reporting numerical margins.
Sources
2 independent publishers whose own reporting we read for this story.
- nature.comPredicting genome-wide functional constraints with GPN-Star
1 article · September 8, 2026
- phys.orgNew AI model for DNA learns from evolution to unlock secrets of the human genome
1 article · September 9, 2026
Topics and entities
Topics
- Variant effect prediction
- Whole-genome alignments
- Efficient model training
- Genomic language models