Genomic Foundation Models

Do genomic foundation models (FMs) provide predictive information beyond simple sequence composition? We evaluate frozen representations from five DNA FMs against normalized k-mer frequencies across 57 genomic classification datasets and two exploratory gene-expression regression tasks. Simple k-mer features remain competitive with most models, whereas NTv3 shows a consistent advantage; the ranking is robust to globally fixed and multi-scale k-mer baselines and alternative downstream probes. To test incremental predictive utility directly, we compare global 6-mer features with their concatenation with each FM embedding under a matched Random Forest protocol. NTv3 improves 46 of 57 datasets (mean $Δ$MCC = +0.072), while Caduceus provides smaller gains and the other models do not improve mean MCC. We separately decompose embeddings into a component linearly predictable from 6-mer frequencies and a ridge residual, finding that linear recoverability does not reliably indicate downstream utility. Exact 6-mer-preserving shuffles further reveal task-relevant sequence-order sensitivity, most pronounced for NTv3, although order sensitivity alone does not guarantee predictive improvements. Our results establish compositional features as essential controls and offer complementary diagnostics for evaluating the added value of frozen genomic representations, without making claims about fine-tuning capacity.

Contemporary Genomic Foundation Models (GFMs) rely on a DNA-as-a-string paradigm that employs masked token prediction objectives for pretraining. However, this abstraction does not explicitly model the biochemical, structural, and physical properties essential to biological function. Many molecular properties can be estimated from sequence using established biophysical models, so their utility lies not in providing an independent modality, but in introducing priors that training objectives can explicitly exploit. We introduce VANDAM, a framework that extends the training of GFMs with DNA molecular priors. In self-supervised training, VANDAM predicts regional molecular properties from pooled representations. When functional labels are available and can reward retaining molecular priors, local features are additionally injected at the input. VANDAM consistently improves downstream performance across four architecture families and nine held-out genomic tasks by complementing token-based objectives. Probing experiments further demonstrate that the use of molecular priors generalizes to other unseen molecular properties.

Genome foundation models are most useful where sequences are generated, yet the largest models need datacenter accelerators and a place to send private DNA. We ask whether a 15-billion-parameter genome mixture-of-experts (MoE) model can instead run on volunteers' web browsers, with the experts spread across many untrusted devices, without changing its predictions and without revealing the sequence to any single device. We build a system in which a trusted coordinator runs attention and routing while browser workers run every expert feed-forward network through hand-written WebGPU kernels, and we protect the expert inputs with real-valued Lagrange coded computing: each worker receives only a Gaussian-padded share, computes the expert's linear maps, and the coordinator decodes from any two of three workers. On GenomeOcean-MoE (8 experts, top-2 routing, 24 layers), the browser path matches native llama.cpp at every quantization level, the distributed path stays at the BF16 numerical noise floor (KL 0.0036 nats per token), and an unfitted latency model predicts decode time within 0.74% (median) under emulated wide-area links. We first show that plaintext expert inputs are not private: a probe recovers the token from a single vector at every depth, and one worker can identify the source genome from 300 unordered tokens with 92% accuracy. With coded experts, an adaptive attacker trained on shares falls to the most-frequent-token baseline, one worker's information about each token is bounded below one bit per forward pass, and the fidelity cost stays below the BF16 noise floor; in Chrome, coded decoding runs at 220 to 376 ms per token, depending on how much of the routing is hidden, and continues without replicas when a worker fails.

A small number of unusually high-gain parameters can exert disproportionate effects in transformer language models, but whether analogous structures recur in genomic foundation models and whether structural geometry determines functional importance remains unknown. We analyzed high-gain rows in gated feed-forward networks across text and genomic foundation models, including a frozen 22-model causal census. Computing an associated bilinear weight operator exactly, without a diagonal approximation, we tested whether structural extremeness is a transferable mechanism. Activation-derived candidates were functionally enriched relative to random and top-norm same-layer controls, yet neither spectral concentration nor operator magnitude predicted causal effect size, and these associations vanished within the endpoint-homogeneous text-decoder subset. A within-layer sweep of 36 rows in one genomic and one text decoder resolved this into two regimes: below the detector's acceptance threshold the ratio carried no positive information about causal damage, whereas above it the ratio ordered rows strongly but did not grade severity as a dose-response. The same sweep revealed a second individually catastrophic row invisible to a one-candidate-per-model census, and non-additive damage among co-located critical rows. Case studies showed divergent causal organizations: a robust super-additive pair interaction in DNABERT-2, and in GENERator a sharply position-localized dependence in which preserving or restoring the row's beginning-of-sequence contribution rescued essentially all native-loss damage. High-gain gated-FFN rows are therefore a recurrent architectural phenotype whose structural prominence acts as an enrichment signal, not a calibrated measure of functional criticality or a specification of causal organization. Enrichment is general, but the mechanism is model-specific.

Bidirectional discrete diffusion model appears naturally suited to genomic modeling because it can reconstruct missing sequence from both flanks. We developed GenDA (Genomic Density-optimized Absorbing Diffusion) under the additional hypothesis that entropy-guided span placement would concentrate reconstruction pressure on compositionally complex regions, improving both downstream variant-effect prediction and functional sequence generation. Our results only partially support this premise. After supervised fine-tuning, the 202M-parameter GenDA model reaches a pooled ClinVar SNV AUROC of 0.774, exceeding a similarly scaled autoregressive model by 0.103. However, a matched random-span variant reaches 0.777, providing no evidence that entropy guidance causes the ClinVar improvement. More unexpectedly, GenDA fails a zero-shot functional inpainting stress test: across promoters, enhancers, exon boundaries, and intron boundaries, it does not consistently outperform a control that shuffles the native gap while exactly preserving 3-mer composition. Failure is already present for 50--500-bp gaps, although enhancer degradation worsens at longer gaps. Diagnostics identify several boundary conditions: entropy measures local sequence complexity rather than functional importance; 1-mer tokenization limits physical context; training spans are capped at 300 bp; and high absolute AlphaGenome fidelity can coexist with negative control-normalized restoration. These results show that strong fine-tuned variant prediction, a plausible corruption prior, and functional generation are distinct claims that require separate validation.

Genomic language models achieve strong performance across regulatory-genomics tasks, yet what these models internally represent remains opaque, and the field lacks a principled procedure for verifying that an apparent concept'' inside a model is real rather than an artifact of sequence composition. We introduce a framework that combines sparse dictionary learning with causal intervention to extract, validate, and causally test interpretable features in genomic foundation models. Training top-$k$ sparse autoencoders on the hidden activations of two architecturally distinct models, Nucleotide Transformer ($6$-mer tokenization) and DNABERT-2 (byte-pair encoding), we recover thousands of monosemantic features that map to transcription-factor (TF) sequence motifs. We show that the naive validation of such features against position weight matrices is severely confounded by GC composition and repetitive elements, producing hundreds of spurious TF features'', and we develop a composition-matched, binding-resolved protocol that removes these confounds. Critically, we move beyond correlation: by ablating individual dictionary directions during the model's forward pass and measuring the induced shift in the model's own predictive distribution, we establish that specific features are causally used to represent cell-type-specific TF binding, not merely motif presence. Across three transcription factors (CTCF, GATA1, REST) and both architectures, causally validated binding features emerge reproducibly ($7$--$14$ of $15$ tested features per condition), while two classes of negative control, scrambled binding labels and randomly selected features, yield no detectable signal. The framework is purely computational, uses only public data, and provides a reusable standard for interpretability claims in genomic deep learning.

Recent breakthroughs in foundation models and Large Language Models (LLMs) have introduced new opportunities for studying and decoding genomic sequences. Several state-of-the-art approaches, such as DNABERT2, rely on transformer-based architectures, while others, such as ConvNova, still build upon more conventional convolutional models. However, systematic benchmark comparisons across these methods remain scarce. Given that transformer-based models require extensive and costly pretraining, it is crucial to evaluate whether their performance gains justify this overhead. Moreover, LLMs such as DNABERT2 typically rely on Byte Pair Encoding (BPE) tokenization, whose relevance for DNA sequence representation is still debated within the genomics community. In this work, we investigate three key questions: (i) do transformer-based models provide sufficient improvements on fine-tuning tasks upon heavy pretraining, (ii) what is the actual contribution of pretraining in this setting, and (iii) how does BPE tokenization impact performance on genomics-related tasks?

High-grade gliomas integrate into neural circuits through functional synapses with neurons, raising the question of which noncoding elements shape synaptogenic gene expression in tumor cells. The regulatory program written across the dark genome, what we call the $\textit{dark regulome}$, is the natural substrate to probe, and sequence foundation models offer a zero-shot route through in-silico mutagenesis (ISM); yet likelihood-based scoring is tautologically coupled to local sequence predictability, leaving the regulatory interpretation underdetermined. Across three architecturally distinct foundation models (Caduceus-Ph, HyenaDNA, Enformer) and 30,448 dark genome elements at 92 glioma-relevant loci, we introduce a residualization-and-permutation diagnostic that separates predictability-driven from regulation-driven RIS variance. A sharp 10kb proximal-regulatory horizon survives every control we apply, but the LM-derived element-class hierarchy does not: a six-feature linear baseline matches Caduceus top-decile membership at AUC $= 0.985$. Cross-architecture decomposition cleanly separates a sequence-predictability layer (the two language models co-rank long well-predicted transposable elements) from a regulatory-output layer (Enformer alone retains residual cCRE-discriminative signal), with literally zero overlap between the two top-100 lists. Conservation, brain cis-eQTL, and STRING-PPI cross-checks then anchor what biology survives: top-100 elements across all three models are $3.3\times$ enriched per model for matching brain eQTLs ($p_\mathrm{emp} < 5\times 10^{-3}$), while a tempting transposable-element regulatory layer and a striking NRXN1+NLGN1 protein-pair convergence both fail proper permutation tests once those tests are constructed. We deliver the diagnostic as a general methodological tool for any ISM-based regulatory study.

Genomic foundation models increasingly adopt large language model architectures, yet almost universally rely on fixed tokenization schemes such as $k$-mers, BPE, or single nucleotides, which impose arbitrary sequence boundaries that may obscure biologically relevant structure. We present LDARNet, a 120M-parameter hierarchical genomic foundation model that adapts H-Net-style dynamic chunking from autoregressive generation to masked language modeling, combining BiMamba-2 state-space layers with local attention, bidirectional routing, and a ratio-based regularizer to induce adaptive token boundaries without supervision. Fine-tuned on 27 tasks from the Nucleotide Transformer and Genomic Benchmarks suites, LDARNet achieves 11/18 wins among compact models ($<$300M parameters) and state-of-the-art results on 5 histone modification tasks, outperforming models up to 20$\times$ larger. A FLOPs-matched controlled experiment isolates learned routing as the source of these gains: learned boundaries beat fixed-grid boundaries by up to 14 percentage points on histone tasks at identical compute. Nucleotide-resolution analysis further shows that the learned boundaries align with canonical promoter motifs and splice junctions without supervision, providing a biological interpretation for adaptive tokenization in genomic foundation models.

Progress in genomic foundation models is difficult to assess due to fragmented benchmarks, incompatible evaluation protocols, and task-specific reporting. As a result, claims of superiority or generality across models are often not directly comparable. We introduce GENEB, a large-scale diagnostic benchmark that evaluates frozen representations from 40 genomic foundation models across 100 tasks spanning 13 functional categories under a unified probing-based protocol, including few-shot regimes. GENEB enables controlled comparison across model scale, architecture, tokenization, and pretraining data while explicitly exposing task-level trade-offs. Our analysis shows that aggregate leaderboards are unstable: model rankings vary sharply across task categories, scale provides only modest and inconsistent gains, and architectural and pretraining alignment frequently outweigh parameter count. These results highlight limitations of current evaluation practices and position GENEB as a reference framework for principled comparison and category-aware model selection in genomic machine learning.

Nucleotide sequences constitute the fundamental genetic basis of biological systems, rendering viral genomic analysis critical for biomedical advancement. Despite progress in biological foundation models, specifically nucleotide foundation models (NFMs), the field lacks a unified standard for viral genomics to facilitate community development and enforce biosecurity constraints. To address this, we introduce ViroBench, the first comprehensive and large-scale benchmark specifically designed for NFMs in viral settings. ViroBench evaluates models across two critical dimensions: biological understanding and latent biosecurity risk, covering 18 diverse scenarios within 4 task types. Extensive evaluation of 66 NFMs across diverse architectures yields three critical conclusions. Firstly, NFMs exhibit a performance degradation in biological understanding under phylogenetic and temporal shifts, indicating weak extrapolation capabilities. Secondly, generation tasks reveal a decoupling between statistical likelihood and biological functional validity, posing latent biosecurity risks. Thirdly, controlled ablation studies reveal that taxonomic diversity in pretraining data outweighs parameter scale. Specifically, a lightweight baseline trained on diverse data achieves a 67.5% performance gain over its original model. Overall, ViroBench provides interpretable, diagnostic evaluations and a reproducible measurement framework for future research on viral nucleotide foundation models. The datasets and code are publicly available at https://github.com/QIANJINYDX/ViroBench.

Foundation models in genomics have shown mixed success compared to their counterparts in natural language processing. Yet, the reasons for their limited effectiveness remain poorly understood. In this work, we investigate the role of entropy as a fundamental factor limiting the capacities of such models to learn from their training data and develop foundational capabilities. We train ensembles of models on text and DNA sequences and analyze their predictions, static embeddings, and empirical Fisher information flow. We show that the high entropy of genomic sequences -- from the point of view of unseen token prediction -- leads to near-uniform output distributions, disagreement across models, and unstable static embeddings, even for models that are matched in architecture, training and data. We then demonstrate that models trained on DNA concentrate Fisher information in embedding layers, seemingly failing to exploit inter-token relationships. Our results suggest that self-supervised training from sequences alone may not be applicable to genomic data, calling into question the assumptions underlying current methodologies for training genomic foundation models.

DNA foundation models have become transformative tools in bioinformatics and healthcare applications. Trained on vast genomic datasets, these models can be used to generate sequence embeddings, dense vector representations that capture complex genomic information. These embeddings are increasingly being shared via Embeddings-as-a-Service (EaaS) frameworks to facilitate downstream tasks, while supposedly protecting the privacy of the underlying raw sequences. However, as this practice becomes more prevalent, the security of these representations is being called into question. This study evaluates the resilience of DNA foundation models to model inversion attacks, whereby adversaries attempt to reconstruct sensitive training data from model outputs. In our study, the model's output for reconstructing the DNA sequence is a zero-shot embedding, which is then fed to a decoder. We evaluated the privacy of three DNA foundation models: DNABERT-2, Evo 2, and Nucleotide Transformer v2 (NTv2). Our results show that per-token embeddings allow near-perfect sequence reconstruction across all models. For mean-pooled embeddings, reconstruction quality degrades as sequence length increases, though it remains substantially above random baselines. Evo 2 and NTv2 prove to be most vulnerable, especially for shorter sequences with reconstruction similarities > 90%, while DNABERT-2's BPE tokenization provides the greatest resilience. We found that the correlation between embedding similarity and sequence similarity was a key predictor of reconstruction success. Our findings emphasize the urgent need for privacy-aware design in genomic foundation models prior to their widespread deployment in EaaS settings. Training code, model weights and evaluation pipeline are released on: https://github.com/not-a-feature/DNA-Embedding-Inversion.

Background: Single-cell foundation models such as Geneformer and scGPT encode rich biological information, but whether this includes causal regulatory logic rather than statistical co-expression remains unclear. Sparse autoencoders (SAEs) can resolve superposition in neural networks by decomposing dense activations into interpretable features, yet they have not been systematically applied to biological foundation models. Results: We trained TopK SAEs on residual stream activations from all layers of Geneformer V2-316M (18 layers, d=1152) and scGPT whole-human (12 layers, d=512), producing atlases of 82525 and 24527 features, respectively. Both atlases confirm massive superposition, with 99.8 percent of features invisible to SVD. Systematic characterization reveals rich biological organization: 29 to 59 percent of features annotate to Gene Ontology, KEGG, Reactome, STRING, or TRRUST, with U-shaped layer profiles reflecting hierarchical abstraction. Features organize into co-activation modules (141 in Geneformer, 76 in scGPT), exhibit causal specificity (median 2.36x), and form cross-layer information highways (63 to 99.8 percent). When tested against genome-scale CRISPRi perturbation data, only 3 of 48 transcription factors (6.2 percent) show regulatory-target-specific feature responses. A multi-tissue control yields marginal improvement (10.4 percent, 5 of 48 TFs), establishing model representations as the bottleneck. Conclusions: These models have internalized organized biological knowledge, including pathway membership, protein interactions, functional modules, and hierarchical abstraction, yet they encode minimal causal regulatory logic. We release both feature atlases as interactive web platforms enabling exploration of more than 107000 features across 30 layers of two leading single-cell foundation models.

Early DNA foundation models adopted BERT-style training, achieving good performance on DNA understanding tasks but lacking generative capabilities. Recent autoregressive models enable DNA generation, but employ left-to-right causal modeling that is suboptimal for DNA where regulatory relationships are inherently bidirectional. We present D3LM (Discrete DNA Diffusion Language Model), which unifies bidirectional representation learning and DNA generation through masked diffusion. D3LM directly adopts the Nucleotide Transformer (NT) v2 architecture but reformulates the training objective as masked diffusion in discrete DNA space, enabling both bidirectional understanding and generation capabilities within a single model. Compared to NT v2 of the same size, D3LM achieves improved performance on understanding tasks. Notably, on regulatory element generation, D3LM achieves an SFID of 10.92, closely approaching real DNA sequences (7.85) and substantially outperforming the previous best result of 29.16 from autoregressive models. Our work suggests diffusion language models as a promising paradigm for unified DNA foundation models. We further present the first systematic study of masked diffusion models in the DNA domain, investigating practical design choices such as tokenization schemes and sampling strategies, thereby providing empirical insights and a solid foundation for future research. D3LM has been released at https://huggingface.co/collections/Hengchang-Liu/d3lm.

When biological foundation models such as scGPT and Geneformer process single-cell gene expression, what geometric and topological structure forms in their internal representations? Is that structure biologically meaningful or a training artifact, and how confident should we be in such claims? We address these questions through autonomous large-scale hypothesis screening: an AI-driven executor-brainstormer loop that proposed, tested, and refined 141 geometric and topological hypotheses across 52 iterations, covering persistent homology, manifold distances, cross-model alignment, community structure, and directed topology, all with explicit null controls and disjoint gene-pool splits. Three principal findings emerge. First, the models learn genuine geometric structure. Gene embedding neighborhoods exhibit non-trivial topology, with persistent homology significant in 11 of 12 transformer layers at p < 0.05 in the weakest domain and 12 of 12 in the other two. A multi-level distance hierarchy shows that manifold-aware metrics outperform Euclidean distance for identifying regulatory gene pairs, and graph community partitions track known transcription factor target relationships. Second, this structure is shared across independently trained models. CCA alignment between scGPT and Geneformer yields canonical correlation of 0.80 and gene retrieval accuracy of 72 percent, yet none of 19 tested methods reliably recover gene-level correspondences. The models agree on the global shape of gene space but not on precise gene placement. Third, the structure is more localized than it first appears. Under stringent null controls applied across all null families, robust signal concentrates in immune tissue, while lung and external lung signals weaken substantially.

Genomic Foundation Models (GFMs) typically rely on Masked Language Modeling (MLM) or Next-Token Prediction (NTP) to learn the "Laws of Nature". While effective at capturing local syntax, these generative paradigms prioritize token-level reconstruction over high-level functional context. We introduce JEPA-DNA, a model-agnostic continual training framework that integrates a Joint-Embedding Predictive Architecture (JEPA) with traditional generative objectives. By supervising global sequence embeddings in a latent space, JEPA-DNA forces models to predict the functional representations of masked genomic segments, shifting the learning signal from token recovery to semantic alignment. We evaluate JEPA-DNA on 17 diverse genomic benchmark tasks, demonstrating consistent gains in linear probing and zero-shot performance regardless of the underlying GFM architecture or generative objective, and establishing a new state-of-the-art for GFMs under the linear-probing protocol. Through extensive ablation studies, we further characterize the synergistic interplay between generative and latent objectives. Our code is provided in the supplementary material.