The challenge that HiFIseek addresses is one of combinatorial chaos. To identify genes whose regulatory regions are enriched for harmful mutations, researchers must first decide which cis-regulatory regions belong to which gene, and then decide how to score the functional impact of the variants found within them. Neither decision is trivial. Dozens of databases and algorithms map regulatory elements to the genes they control, drawing on chromatin contact data, expression quantitative trait loci, and evolutionary conservation. Dozens more methods assign functional impact scores to variants, each trained on different data and embodying different assumptions about what makes a non-coding change deleterious. The Norwegian-led team discovered that when they compared these methods head to head, the agreement between them was strikingly poor, meaning that two researchers analyzing the same mutation dataset could reach entirely different conclusions depending purely on which tools they happened to pick.

This lack of consensus is more than a technical inconvenience. It means that many published analyses of non-coding variation may have been inadvertently shaped, or even driven, by methodological choices rather than by biology. A gene flagged as harboring dangerous regulatory mutations under one combination of gene-region mapping and impact scoring might vanish entirely under another. The team behind HiFIseek argues that the field needs a unified framework in which many such combinations can be evaluated simultaneously, so that findings robust across methods can be distinguished from artifacts of any single pipeline. HiFIseek was built precisely for this purpose, implemented as a reproducible Snakemake workflow that systematically explores the full space of region-score combinations rather than committing to one.

At its core, HiFIseek performs enrichment testing. For each gene, the pipeline gathers all the cis-regulatory regions associated with it under a given mapping method, collects the functional impact scores of the variants falling within those regions, and asks whether high-impact variants occur more frequently than chance would predict. The tool implements two complementary enrichment detection methods, allowing users to cross-validate their findings. By running this procedure across every possible pairing of gene-region association resources and functional impact scoring schemes, HiFIseek produces a consensus picture: genes that light up consistently across many combinations earn confidence, while genes that appear under only one combination are treated with appropriate skepticism.

To demonstrate the tool’s power, the researchers applied it to eight different cancer cohorts, searching for genes whose associated regulatory regions showed enrichment of high functional impact cis-regulatory variants. The results were revealing. Cancer driver genes, the well-known engines of tumor development, exhibited a higher frequency of damaging regulatory mutations compared to non-cancer genes, supporting the idea that tumors exploit regulatory disruption as a route to malignancy just as they exploit coding mutations. Even more striking, despite the limited agreement across individual region-score combinations, the genes detected by HiFIseek in the cancer cohorts consistently pointed toward known cancer-related pathways, suggesting that the tool captures genuine biological signal even when its component methods disagree.

The team also turned HiFIseek loose on a set of breast cancer risk-associated single-nucleotide polymorphisms, the common variants identified by genome-wide association studies that have long frustrated researchers because most sit in non-coding regions of unknown function. Here again, consensus across combinations was limited, but one particular combination of mapping and scoring methods returned a list of 18 known cancer-related genes, including BRCA1, the most famous breast cancer gene of all. These genes showed enrichment of high-impact cis-regulatory variants, providing a mechanistic explanation for why variants in distant, apparently inert stretches of DNA might elevate breast cancer risk: they may be tampering with the regulatory switches that control the expression of genes whose loss or gain promotes tumor formation.

The implications for cancer genomics are considerable. Genome-wide association studies have identified hundreds of loci linked to cancer susceptibility, but translating those statistical associations into biological understanding has been slow, precisely because the causal variants and the genes they affect remain obscure. HiFIseek offers a principled way to prioritize candidates, ranking genes by the accumulated evidence that their regulatory apparatus is under mutational attack. In clinical settings, the same logic could eventually help interpret the non-coding variants that flood out of tumor sequencing pipelines, most of which are currently dismissed as variants of uncertain significance because they fall outside coding regions.

Equally important is the study’s methodological warning. By demonstrating that gene-region association resources and functional impact scoring methods exhibit little consensus, the authors have quantified a source of irreproducibility that the field has largely ignored. Their finding that the choice of a specific combination of these tools materially changes which genes are predicted to be cis-regulatory altered should prompt a re-examination of past studies and a more cautious interpretation of future ones. The solution they propose, running all combinations within a single reproducible pipeline, transforms what was previously a hidden source of bias into an explicit dimension of analysis, one that can be interrogated and reported transparently.

HiFIseek’s design also reflects broader trends in computational biology toward reproducibility and modularity. Built on Snakemake, a popular workflow management system, the pipeline can be rerun, extended, and audited by other researchers, addressing the crisis of irreproducible bioinformatics analyses. As new gene-region association datasets and impact scoring methods continue to appear, they can be slotted into the framework, allowing the consensus landscape to be updated continuously. The authors suggest that this systematic, combination-aware approach could serve as a template for other problems in genomics where multiple competing methods exist and their interactions have never been formally assessed.

For now, HiFIseek stands as both a practical discovery tool and a conceptual corrective. It shows that the non-coding genome, long treated as the neglected countryside surrounding the bustling cities of protein-coding genes, harbors mutations with real power to drive disease, and that finding them requires looking through many lenses at once. As tumor sequencing becomes routine and genome-wide association studies grow ever larger, tools like HiFIseek will be essential for converting the noise of non-coding variation into the signal of dysregulated genes, bringing the dark matter of the genome one step closer to the clinic.

Subject of Research: Computational detection of enrichment of high-impact cis-regulatory mutations in gene-associated genomic regions in cancer

Article Title: HiFIseek: Gene-specific enrichment of high-impact mutations in associated genomic regions

Article References: Riudavets Puig, R., O’Mahony, D. G., Brorson, I. S., Sokolova, K., Kristensen, V. N., & Mathelier, A. (2026). HiFIseek: Gene-specific enrichment of high-impact mutations in associated genomic regions. PLOS Computational Biology, 22(10), e1014842. https://doi.org/10.1371/journal.pcbi.1014842

Image Credits: AI Generated

DOI: 10.1371/journal.pcbi.1014842

Keywords: HiFIseek, cis-regulatory variants, non-coding mutations, cancer genomics, functional impact scores, gene regulation, BRCA1, breast cancer, Snakemake pipeline, enrichment analysis, genome-wide association studies, bioinformatics

Predicting how a person’s unique genetic makeup shapes their gene expression has long been a central goal of modern biology. Expression quantitative trait locus, or eQTL, studies have linked thousands of genetic locations to gene activity, but these statistical approaches depend on massive population datasets and are therefore largely restricted to common genetic variants. Rare and novel variants, which are often critical for understanding disease mechanisms, remain stubbornly difficult to interpret. Sequence-to-function models such as Enformer and Borzoi were designed to close this gap by using deep learning to map DNA sequence directly to regulatory outputs, and they perform well when predicting average expression across genes and tissues. Yet a series of benchmarking studies published in recent years showed that these models capture very little of the transcriptomic variation between individuals, prompting researchers to ask whether a fundamentally different approach was needed.

The Hong Kong team, led by Shumin Li, Ruibang Luo, and Yuanhua Huang, saw promise in a newer class of models: genomic language models, or gLMs. Unlike sequence-to-function models, which are trained on curated functional annotations across many cellular contexts, gLMs such as Evo2-7B, the Nucleotide Transformer, and Caduceus learn generalizable representations of DNA through self-supervised pretraining on raw sequence alone. Whether these models could complement or extend existing sequence-to-function approaches for personal expression prediction was an open question. xDecoder was built to answer it. The framework is a lightweight convolutional neural network decoding tower that consumes the full token-level embeddings produced by six different foundation models, three genomic language models and three sequence-to-function models, and translates them into personal gene expression predictions.

The researchers trained and evaluated xDecoder using paired genome and transcriptome data from the GEUVADIS consortium, which includes RNA sequencing and high-quality genotypes from lymphoblastoid cell lines of 1000 Genomes Project participants. Individualized DNA sequences were reconstructed by substituting single-nucleotide variants from phased genotypes into the reference genome. The team focused on genes with significant cis-eQTLs, reserving 200 genes for personalized training and holding out 287 genes on chromosomes 5 and 10 for stringent, locus-disjoint evaluation, with 50 individuals for training and 100 for testing. Importantly, the xDecoder architecture deliberately avoids compressing embeddings through global pooling. Because only about 0.01 percent of positions differ between any two people across a gene’s sequence, preserving positional resolution is essential, so the model aggregates local neighborhoods with small stacked depthwise-separable convolutional layers before pooling and regression, and averages predictions across the two haplotypes.

The study’s central finding is a tale of two learning regimes. In the few-shot setting, where models are fine-tuned on paired personal genome-expression data for genes seen during training, xDecoder delivered consistent and substantial improvements. Cross-individual prediction correlations more than tripled relative to pretrained Enformer and Borzoi and to reference-trained xDecoder variants, rising from 0.089 to 0.194 on average. For specific genes the gains were striking: for HLA-DQA1, pretrained Enformer and AlphaGenome produced negative correlations, while individually fine-tuned xDecoder models exceeded 0.3, with a Caduceus-based variant reaching 0.74. The best genomic language model embeddings even outperformed the best sequence-to-function embeddings, and rp-Caduceus successfully modeled 46 genes for which the widely used linear method PrediXcan produced identical, uninformative predictions across all individuals. Exploratory variant-level perturbation analysis further showed that high-scoring xDecoder perturbations overlapped known eQTL variants, hinting that the model can nominate candidate cis-regulatory variants in loci where fixed genotype weights fail.

The zero-shot picture was far less encouraging. When the models were asked to predict expression across unseen individuals for genes never encountered during training, none performed reliably. Zero-shot performance remained close to zero even as the number of training individuals was scaled from 50 up to 250, demonstrating that the limitation is not simply a matter of insufficient data. Performance did improve modestly for genes with stronger cis-genetic signal, particularly for AlphaGenome, but even in the most favorable strata, zero-shot results fell well below gene-specific supervised models such as PrediXcan, which achieved a mean Spearman correlation of 0.216. The authors characterize this as a persistent cross-locus transfer bottleneck: current sequence models fail to learn a transferable grammar of cis-regulation that generalizes from one genomic neighborhood to another. Cross-population experiments added a further caution, showing that benefits of individual-level training diminished when models trained on European individuals were tested on Yoruba individuals from Ibadan, with genes carrying higher novel variant rates and less similar linkage disequilibrium structure showing larger performance drops.

To diagnose what information DNA-only models are missing, the team conducted a revealing experiment with chromatin accessibility. By adding a base-resolution ATAC-seq channel to Caduceus embeddings, they significantly improved prediction for unseen genes, lifting the mean cross-individual correlation from -0.003 to 0.129 and flipping correlation direction to positive for many genes; for the gene CHST3, adding the ATAC track transformed a wrongly signed prediction into a correlation of 0.755. However, control analyses complicated the interpretation. In the unseen-gene setting, a simple unsupervised baseline that just summed ATAC signal across the same genomic window was competitive with the full model, suggesting that measured chromatin accessibility largely explains the gain. Meanwhile, AlphaGenome-predicted ATAC signals showed only modest concordance with observed accessibility across individuals, indicating that current DNA-only models cannot yet reliably infer a person’s chromatin state from sequence alone. In the seen-gene setting, by contrast, the full DNA-plus-ATAC model outperformed an ATAC-only linear baseline, showing that sequence-derived information adds genuine value when paired training data exist.

The authors are careful about the limits of their advance. Absolute prediction accuracy, while improved, remains modest, and the framework should not yet be interpreted as reliable enough for gene-level prediction from personal genomes in any clinical sense. The evaluation was restricted to lymphoblastoid cell lines, and extending the approach to larger cohorts, multiple tissues, and context-specific regulation remains future work. The researchers also stress that variant prioritization applications require systematic benchmarking against eQTL-supported variants, and that clinical interpretation would demand far stronger calibration and independent validation. Rather than presenting DNA-plus-ATAC inputs as a practical replacement for RNA sequencing, they frame the ATAC experiments as a diagnostic that pinpoints the key bottleneck of current DNA-only models.

Even so, the study sketches a concrete roadmap for the next generation of personalized regulatory models. Combining pretrained sequence encoders with variant-aware modules, using individual-level multi-omic profiles such as ATAC-seq as auxiliary supervision or intermediate regulatory-state targets, and exploiting the few-shot regime where sequence-based models already outperform fixed-variant statistical tools together form a coherent research agenda. Because sequence-based models operate directly on DNA, they can in principle evaluate perturbations outside the fixed variant-weight space of linear predictors, offering potential utility wherever rare or novel variants lack population-derived weights. The xDecoder code has been released openly on GitHub, and the authors suggest that bridging the cross-locus generalization gap, likely through multi-omic, variant-aware training, is the defining challenge that will determine whether genomic foundation models can ultimately deliver truly personalized expression prediction from DNA alone.

Subject of Research: Few-shot personal gene expression prediction using genomic foundation models

Article Title: xDecoder unlocks the potential of genomic foundation models for few-shot personal gene expression prediction

Article References: Li, S., Luo, R., & Huang, Y. (2026). xDecoder unlocks the potential of genomic foundation models for few-shot personal gene expression prediction. Molecular Systems Biology. https://doi.org/10.1038/s44320-026-00238-1

Image Credits: AI Generated

DOI: 10.1038/s44320-026-00238-1

Keywords: genomic foundation models, gene expression prediction, personalized genomics, xDecoder, genomic language models, sequence-to-function models, eQTL, chromatin accessibility, ATAC-seq, deep learning, GEUVADIS, cis-regulatory variants