Papers

Genomic and metagenomic analyses play critical roles in many fields, such as precision medicine, urgent clinical settings, discovering early warnings of communicable diseases, ensuring food safety through pathogen monitoring, agriculture, and scientific discovery. Due to the challenges of analyzing and storing massive volumes of genomic and metagenomic sequence data, significant efforts have been made to accelerate (meta)genomic analyses and store sequence data compressed. Despite the benefits of these techniques, we identify two major outstanding problems in accessing stored sequence data and supplying it to the analysis units: (i) the data movement bottleneck due to moving large amounts of low-reuse data from storage and the unnecessary burden on the rest of the system, and (ii) the data preparation bottleneck, where compressed sequence data needs to be first decompressed and formatted before analysis. In this dissertation, we present customized storage-centric systems, which efficiently (i) analyze (meta)genomic data inside the storage system, and (ii) enable highly-compressed storage and high-performance access of large-scale sequence data, thereby alleviating the overheads of data movement, computation, and data preparation. We demonstrate that the proposed systems significantly improve system performance, energy efficiency, and system cost-efficiency of (meta)genomic analysis. We hope that the storage-centric systems proposed in this dissertation facilitate the broader adoption of (meta)genomic analyses and inspire future research to fundamentally improve the performance, energy efficiency, and cost-effectiveness of other data-intensive application domains related to health and life sciences.

Spatial transcriptomics (ST) provides spatially resolved gene expression profiling but remains expensive, motivating the prediction of ST from histology images. Generative models have emerged as a mainstream paradigm for ST prediction due to their ability to model the conditional distribution of gene expression and capture its inherent stochasticity. However, these methods typically treat genes as independent prediction targets and overlook the intrinsic gene-gene interactions in biological systems, which limits their ability to preserve biologically meaningful co-expression patterns. We argue that gene-gene interactions, which reflect shared pathways and regulatory mechanisms, are essential for generating numerically accurate and biologically coherent ST profiles. In this paper, we propose CorrFlow, a correlation-guided flow matching framework for histology-to-ST prediction that explicitly models gene-gene dependencies through two complementary mechanisms. First, we introduce an annealed masked flow matching strategy, where subsets of genes are progressively masked following a timestep-dependent annealing schedule, encouraging the model to infer masked genes conditioned on the remaining genes and promoting joint conditional modeling beyond per-gene marginal estimation. Second, we devise a gene graph-regularized optimization scheme that integrates prior knowledge from the STRING database and data-driven co-expression estimated by WGCNA to construct a gene affinity graph, which enforces both local consistency and global smoothness in the predicted expression. Extensive experiments across 12 datasets show that CorrFlow achieves the best average PCC and HPCC among evaluated methods, leading to more biologically coherent ST predictions.

Batch integration is a central preprocessing step in single-cell genomics, where datasets collected across experiments, donors, and protocols must be combined despite pervasive technical batch effects. The leading integration methods produce a shared low-dimensional embedding, which discards the corrected gene-expression values that downstream differential-expression, biomarker, and other analyses depend on. The feature-space (expression-correcting) methods that do preserve genes typically align the marginal expression distributions of batches and thereby risk erasing genuine biological variation whenever cell-type composition differs across batches, i.e. whenever batch effect and biological signal are confounded. We recast single-cell batch integration as confounded domain adaptation and apply ConDo, a method that matches conditional expression distributions given the cell-type annotation rather than marginal distributions. To extend ConDo's pairwise source-to-target adapter to the many-batch setting, we introduce an agglomerative compatibility-graph integrator: batches are nodes connected when they share a cell type, and we greedily merge each best-scoring neighbor into a growing reference by fitting one ConDo adapter. On the Open Problems Batch Integration Benchmark, ConDo is the strongest feature-space integrator, ranking first among feature methods on five of six datasets. Furthermore, it is competitive with or better than deep embedding methods on the overall score, ranking first across all methods on four of six datasets while returning corrected expression rather than an opaque embedding.

In the tumor microenvironment, a cell's state is influenced by cell-cell interactions (CCIs) with neighboring cells in its niches. Identifying dysregulated CCIs that are associated with pathogenic processes pinpoints targets for drug discovery. Imaging-based spatial transcriptomics and single-cell RNA sequencing provide, respectively, single-cell spatial information and transcriptome-wide measurements needed to study CCIs, but neither modality provides both. Existing spatial transcriptomics foundation models also cannot effectively learn from spatially resolved single-cell data with full-transcriptome coverage, explicitly infer the CCI mechanisms driving cell state-niche associations, or be interpretable enough to support direct biological interpretations. Here, we present GITIII-scale, a hierarchical, interpretable pan-cancer spatial transcriptomics foundation model for TME representation learning that investigates cell state-niche associations and their underlying ligand-receptor (LR) signaling pathways. GITIII-scale uses transformers to model interactions between pairs of cells at defined spatial distances, an interpretable single-layer graph transformer without a feed-forward network to decompose how each gene in a receiver cell is influenced by each neighboring sender cell, and a graph transformer to generate cellular-neighborhood embeddings. Trained on our assembled pan-cancer database of specimen-matched scRNA-seq and imaging-based spatial transcriptomics datasets, GITIII-scale generated TME embeddings that recovered niche-associated state changes more accurately than existing spatial transcriptomics foundation models in cancer types unseen during training. A case study of an unseen breast cancer dataset further demonstrated the model's interpretability by identifying potentially drug-targetable LR pathways associated with endothelial overgrowth and tumorigenesis.

Messenger RNA (mRNA)-based therapeutics have emerged as a powerful platform for vaccines, protein replacement therapies, and cancer immunotherapy. A critical bottleneck in mRNA development is manufacturing large quantities of RNA economically, as measured by RNA yield emerging from an in vitro transcription (IVT) reaction. However, how promoter-adjacent DNA sequences influence RNA yield remains poorly characterized. Here, we present an integrated deep learning framework that leverages massively parallel next-generation sequencing (NGS) assays to measure RNA yield across large sequence spaces. A library of 10^5 randomized oligonucleotide sequences was designed to systematically explore sequence diversity within a defined structural context. DNA and RNA abundances were quantified in parallel using Illumina sequencing, enabling high-resolution measurement of sequence-to-yield relationships at scale. Sequences were one-hot encoded and used to train deep learning models, using a convolutional neural network architecture. The model achieved a Pearson correlation of 0.94 between predicted and experimentally measured RNA yield on a held-out test set, demonstrating strong generalization across diverse sequence contexts. Importantly, the trained model can be deployed in a production environment to score and rank novel RNA sequence designs by predicted IVT yield, enabling cost-effective, pre-experimental prioritization of the most manufacturable candidates. This framework establishes a scalable, data-driven approach to DNA and RNA sequence optimization, with broad applicability to vaccine antigen design, therapeutic protein delivery, and synthetic biology. By integrating high-throughput experimentation with advanced deep learning modeling, it significantly reduces screening costs and accelerates RNA engineering cycle times.

Full-length RNAs, particularly messenger RNAs, often exceed the context lengths used to pretrain existing RNA foundation models, limiting complete-transcript modeling at single-nucleotide resolution. We present RIBOSPAN, a 1.61-billion-parameter bidirectional RNA foundation model natively pretrained with context lengths up to 10,240 nt. RIBOSPAN combines dense bidirectional self-attention, single-nucleotide tokenization, and attention-isolated sequence packing to enable high-resolution modeling of complete long RNAs. Native 10K pretraining preserves strong reconstruction at 10,240 tokens and, in a controlled long-context benchmark, maintains strong contextual responsiveness and context-specific representation separation while keeping perturbation-induced changes highly localized. Inference-time YaRN scaling recovers much of the contextual organization lost by direct short-context extrapolation, but induces substantially greater distal representation diffusion. Frozen RNA-type evaluations show that RIBOSPAN learns state-of-the-art RNA representations, with a particularly clear advantage on long RNAs. Across downstream biological benchmarks, RIBOSPAN emerges as the strongest encoder-only RNA foundation model, achieving state-of-the-art performance in both full-transcript biological property prediction and zero-shot mutation-fitness modeling. Building on the same backbone, we develop a multidimensionally conditioned discrete-diffusion framework for full-length mRNA generation and redesign, including synonymous-codon diffusion for protein-preserving CDS optimization. Together, RIBOSPAN establishes a powerful long-context foundation for transferable RNA representation learning, biological prediction, and full-transcript mRNA design.

The 5' untranslated region is a major determinant of translation initiation, and its effect becomes especially important in modified mRNA sequences, where start-codon context, cap-proximal secondary structure, upstream AUGs and upstream open reading frames, and nucleotide chemistry can alter ribosome scanning and initiation recruitment, scanning, and decoding in sequence-dependent ways. Recent computational studies have moved the field from prediction toward design, including massively trained predictive models such as Smart5UTR for m1$\Psi$-modified mRNA, broader 5' UTR generation and optimization frameworks such as UTRGAN and UTailoR, and structure-guided RNA design systems such as RhoDesign. Here, we describe a conditional generative framework for 50-nt modified-RNA 5' UTR design that optionally conditions on ribosome load, GC content, minimum free energy, and target secondary structure. The implementation uses a Transformer-based generator followed by sequence ranking and local refinement with a Smart5UTR-derived ribosome-load oracle and ViennaRNA-based folding metrics, including support for modified-base folding parameters. Across multiple simulation scenarios and experimental settings, different combinations of RL, GC, MFE, and structural constraints produced distinct performance tradeoffs, enabling ablation-based identification of the best-performing formulation.

Inferring continuous system evolution from sparse temporal snapshots is a key challenge in generative modeling and single-cell omics. While Optimal Transport (OT) is popular, existing frameworks are largely restricted to first-order dynamics, assuming memoryless velocity fields. This limits expressiveness, as first-order systems fail to account for regulatory momentum and time-delayed responses inherent in processes like cell differentiation. Here, we introduce TracingFlow, a simulation-free Flow Matching framework generalizing to second-order dynamics. By using neural networks to regress the acceleration field, TracingFlow provides an exact, efficient solution to the Dynamical Optimal Acceleration Transport (DOAT) problem. Unlike first-order methods yielding over-smoothed trajectories, our second-order formulation captures high-curvature transitions and nonlinear evolutions by learning the underlying force fields. Evaluated on complex synthetic and large-scale scRNA-seq datasets, TracingFlow achieves superior accuracy in distributional reconstruction and trajectory faithfulness. Moreover, by integrating lineage tracing priors, it recovers dynamical structures that are both mathematically optimal and biologically plausible.

Gapped k-mer support vector machines (gkm-SVMs) are widely used for classifying regulatory DNA sequences and identifying the sequence features underlying those predictions. Although LS-GKM provides an efficient implementation of gkm-based kernels, it is restricted to binary classification and does not provide calibrated probability outputs. Here, we present mLS-GKM, an extension of LS-GKM that adds multiclass classification, probability-calibrated predictions, parallelised inference, memory efficient sequence interpretation and checkpointing during training. Across 322 ENCODE ChIP-seq datasets, classifiers trained using mLS-GKM are identical to those produced by LS-GKM, while gkmpredict and gkmexplain run 22x and 80x faster respectively at 64 threads, and gkmexplain peak memory is reduced by more than 65%. As a practical demonstration, mLS-GKM was able to accurately distinguish between enhancers, promoters and CTCF binding sites directly from sequence and identified biologically relevant regulatory motifs. Together, these improvements extend gkm-SVMs to multiclass problems and substantially improve their scalability, enabling efficient interpretation of regulatory DNA sequences in large and complex datasets

DNA methylation provides a stable record of cellular identity, capturing epigenetic programs that distinguish specialized cell states despite a shared genome. Because malignant transformation and tumour progression are accompanied by extensive epigenetic remodeling, we hypothesized that the methylome of melanocytic lesions contains biologically and clinically relevant information for both diagnosis and disease progression. In a cohort of 1,001 tissue samples prospectively collected across eight German university hospitals profiled using Illumina Infinium MethylationEPIC arrays, we compared machine-learning models based on selected Cytosine phosphate Guanine (CpG) methylation sites with models incorporating biology-guided features, including epigenetic age acceleration, cell type composition and copy-number variation burden. In an external test set, the best diagnostic classifier was CpG-based and distinguished melanocytic nevi, noninvasive melanoma and invasive melanoma with a macro-averaged area under the receiver operating characteristic curve of 0.919 (95% CI: 0.878 to 0.952). Notably, across CpGs most strongly hyper- and hypomethylated between NV and IM, NIM showed an intermediate methylation profile, providing a molecular correlate of its diagnostic complexity. The best model for clinically relevant treatment group prediction, with AJCC stages grouped according to guideline-based management recommendations, relied on biology-guided features and achieved a macro-averaged mean absolute error of 0.627 (95% CI: 0.477 to 0.808). Together, these findings demonstrate that methylation-based models can capture both diagnostic identity and clinically relevant disease stratification, supporting DNA methylation as a promising biomarker for further validation and potential clinical translation.

We study whether the loss design of High-Fidelity Generative Image Compression (HiFiC), a GAN-based neural codec originally built for natural photographs, can be adapted to preserve biologically meaningful structure in Hi-C chromatin contact maps under lossy compression. Standard image compression, including HiFiC in its original form, optimizes for human visual perception; but a Hi-C contact map is normally distributed together with its numeric matrix file (.cool/.mcool), which downstream genomic analysis tools consume directly. Aggressive compression that looks acceptable to the eye can nonetheless blur or delete loops and topologically associating domain (TAD) boundaries that these tools depend on. We modify HiFiC's distortion term with a spatially-weighted MSE that up-weights biologically salient regions (loops, TAD boundaries, stripes, compartment structure) and add an insulation-score loss term that directly penalizes loss of TAD boundary sharpness. We describe a three-phase fine-tuning strategy that adapts a pretrained HiFiC checkpoint to the Hi-C domain without catastrophic forgetting. We evaluate the resulting system, HiFiC-G, using both conventional image-quality metrics (PSNR, SSIM) and genomics-domain preservation metrics (loop/TAD/compartment/stripe preservation percentage) across two cell lines. HiFiC-G preserves local structure, meaning stripes and TAD boundaries, substantially better than the metrics alone would suggest, while long-range A/B compartment structure remains poorly preserved; we show this gap tracks genomic scale and is consistent with a specific architectural cause, the fixed-size tiling that both HiFiC-G and the original HiFiC rely on for memory efficiency.

This study aims to explore the applicability of Transformer-based models for genetic sequence classification by evaluating their performance in predicting hepatitis C virus (HCV) genotypes and subtypes after fine-tuning. A total of 2,881 HCV whole-genome sequences obtained from the Los Alamos HCV Sequence Database were used, including genotypes 1 to 6 and all confirmed subtypes. Genotypes 7 and 8 were excluded due to an insufficient number of samples. The fine-tuning process was based on several datasets that differed in fragmentation method, data volume per file, and labeling. In genotype classification, fine-tuning strategies employing homogeneous fragmentation and balanced sample distribution resulted in higher performance, with precision ranging from 98.48% to 100%. In contrast, fine-tuning conducted using a fragmentation strategy that caused data imbalance, along with an arbitrary distribution of samples across training files, achieved a precision of 48.12%, which is considered low compared with other models. This configuration, which was also manually evaluated, resulted in a high error rate in genotype 5 prediction due to its low frequency in the datasets used. In subtype classification, the best-performing fine-tuning approach achieved 99.89% accuracy and 99.87% precision. Models that included additional genotypes showed a slight decrease in performance due to the increased complexity of the task. This study demonstrates that, when fine-tuning datasets contain properly fragmented, distributed, and labeled genetic sequences, Transformer-based neural networks can achieve high performance and are a promising approach for HCV genotype and subtype classification.

Deep clustering models for single-cell RNA sequencing often assign cells through latent or centroid-based mechanisms that are difficult to inspect. We introduce scDNM-VAE (single-cell Dendritic Neuron Model Variational Autoencoder), a deep clustering framework that combines a variational autoencoder with a dendritic neuron-inspired head. Cluster assignments are governed by learnable signed synaptic weights and thresholds: the weight sign determines the direction of a gate's response to a latent coordinate, its magnitude controls steepness, and the weight-threshold pair determines the transition location. The trained clustering function can therefore be inspected directly without fitting a post-hoc explanation model. We benchmark scDNM-VAE on four datasets spanning immune, cortical, cardiac, and hematopoietic cells against scVI followed by KMeans and an MLP-DEC ablation. scDNM-VAE performs better than scVI on PBMC3k, comparably on the Human Heart Cell Atlas and Paul15, and worse on Zeisel, while producing biologically coherent marker-gene signatures. Ablating each cluster's three highest-magnitude synaptic dimensions causes numerically greater reassignment than random-dimension ablation across all datasets, but the margins are modest and negligible on Zeisel. These results show that signed dendritic gating supports competitive clustering with a parameter-inspectable decision function, while indicating that decision-relevant information is distributed across the latent space.

Background. Artificial-intelligence design tools now plan genome-scale edits, and agentic systems execute those plans with progressively less human oversight. Biosecurity controls are limited to two points: refusal guardrails at the foundation model and sequence-identity screening at the synthesiser. The design stage between them, where the plan is specified, remains governed by recommendations rather than any deployed system. Results. We present BioFirewall, a rule-governed middleware that intercepts a genome-writing plan and returns allow, flag-for-review, or refuse across five hazard axes native to genome writing: cargo, locus, edit type, germline and scale, with cited evidence, a signed design passport, a tamper-evident audit log, and tiered access. On a de-circularised benchmark of safe proxies scored against independent oracles, a function-aware cargo classifier reached a true-positive rate of 0.72 (95% CI 0.43 to 0.89) at a 1% false-positive rate, whereas frontier and open language-model judges did not screen the same sequences reliably. Under prompt injection, the open-weight judges flipped their blocking verdict to allow in 3 and 5 of 6 trials per channel, while the deterministic screen remained invariant. None of 288 legitimate plans from three templates was refused, yielding a certified 95% upper bound of 0.0103 on the false-refuse rate, and a session monitor intercepted cross-call decomposition attacks. On a held-out gene set, the locus axis was enriched for drivers of in vivo insertional oncogenesis (AUROC 0.605; odds ratio 3.34). Conclusions. Design-stage governance is achievable in practice. BioFirewall is released as open source with a pre-registered, open-data-reproducible benchmark.

Background. Language-model agents are widely used in biology, but they report quantities without a verifiable source and pose unmanaged biosecurity risks. Genome writing sharpens both: a write plan must specify a location, writer enzyme, cargo, and delivery vehicle, all quantitative and interdependent, so without an integrated tool layer, the agent must supply them. We introduce PEN-STACK, an open tool layer that supplies them with guaranteed provenance. Results. PEN-STACK provides ten genome-writing design stages as twenty-two scope-aware tools, accessible via a software development kit, a Model Context Protocol server, and a Representational State Transfer interface, under a type-enforced invariant: every quantity must originate from a validated tool. Without tools, three model families fabricated 90.8% to 98.8% of the 240 required quantities under a naive prompt; coaching left a residual of 0 to 4, with no model certified at zero. Driving the tools, the same models fabricated nothing on a four-goal audit. A pre-emission biosecurity screen matched expert labels on all eight designs. The expression-robustness axis validated at exact-site resolution (\rho = 0.571, n = 1,506) but not at the coarser resolution served by default (\rho \approx 0.16), which returns a machine-readable downgrade flag. Eight of ten pre-registered claims did not pass, each flagged as machine-readable. Conclusions. On this evidence, grounding, not prompting or model scale, removes fabrication, and grounding requires a substrate; the grounded arm, a four-goal audit, warrants replication at the 240-field scale. PEN-STACK provides that substrate as open, importable code for agentic genome-engineering systems.

Bioinformatics workflows rely heavily on visual representations. Quality-control plots, cell embeddings, heatmaps, genome-browser tracks, and interactive dashboards are not merely illustrations, but instruments for making analytical decisions. For blind and low-vision researchers who use screen readers, braille displays, or audio-based interfaces, these create a barrier: the evidence used to justify an analysis is often encoded in visual form, while the underlying decision remains undocumented. We argue that non-visual accessibility and computational reproducibility are closely aligned, as they both require analyses to be transparent and to record why decisions were made. We present ten simple rules for non-visual bioinformatics, covering plots as decision records, cautious use of AI-generated figure descriptions, accessible computing environments, text-first literate programming, structured data and metadata, compact object summaries, accessible publication formats, collaboration practices, shared community infrastructure, and accessibility as part of FAIR research. The intended audience is computational biologists and developers. Using single-cell RNA-seq as a running example, we show that the accessible equivalent of a plot is a structured decision record. That is, a plot companion that goes beyond storing the underlying data by also stating the purpose of the analysis and the resulting quantitative evidence and uncertainty. We argue that treating accessibility in this way makes bioinformatics more inclusive and also more transparent and auditable.

The field of bioinformatics struggles with legacy code - old code that is commonly used but may no longer have a maintainer, or may be written in an now-unfamiliar language (e.g. Perl, Fortran). This incurs maintenance cost (technical debt), but dynamically typed languages also negatively impacts the environment and fail to make use of modern hardware. Legacy code may also have security or safety problems that make it unsuited for use in clinical settings. Here we show that agentic AI, combined with static analysis, can be used to translate legacy code to the modern language Rust. We provide prompts and supporting software to aid systematic translation, and evaluate it on common software for NGS and imaging. We showcase the result on our software Bascet: Size was reduced by ~80x, build time decreased by ~10x, and performance of key steps improved >3x. Unix dependencies were also removed, making Bascet the only single-cell pipeline able to run on native Windows, without a container. Large-scale refactoring of bioinformatics software is thus now possible at a limited budget, enabling more complex tools to be developed.

SVPLEX is a Nextflow pipeline for cohort-level structural variant detection from short-read whole-genome sequencing data. The pipeline implements six different structural variant callers with different strengths and weaknesses, integrating different levels of evidence for SVs, and generates a merged consensus callset across the analysis cohort. Callset filtering is achieved by leveraging consensus among multiple individual callers and by ensuring that deletion and duplication calls are supported by observable changes in read depth. The output merged cohort SV callset can then be used to assess cohort-specific variation, remove technical artefacts, and serve as input for rare disease variant prioritisation workflows. SVPLEX is user-friendly, reproducible, scalable, and can be executed flexibly on either a local workstation, a high-performance compute (HPC) cluster, or deployed on cloud infrastructure. The required inputs are alignment files for the cohort of interest, and the output is a single merged cohort structural variant VCF. SVPLEX is available on GitHub (bahlolab/SVPLEX) and is licensed under the MIT open-source licence.

Omics datasets, particularly single-cell RNA sequencing data, are high-dimensional, sparse, noisy, and dominated by zero values, making faithful low-dimensional representation challenging. Existing dimensionality-reduction methods may distort local neighbourhoods, global organization, or the cohesion of meaningful populations, with similar limitations arising in genealogical data. We introduce Contrastive Manifold Approximation and Projection (CosMAP), a graph-based unsupervised dimensionality-reduction method for producing faithful and interpretable embeddings. CosMAP extends the graph-based framework of UMAP by combining cosine-similarity neighbourhoods with temperature-normalized contrastive affinities, which are optimized in the embedding space using an attractive–repulsive objective. It further employs a two-phase refinement strategy: an intermediate higher-dimensional representation is first learned and then used to reconstruct the neighbourhood graph and initialize the final low-dimensional embedding. We evaluate CosMAP on MNIST and USPS handwritten-digit datasets, mouse retina and cortex single-cell RNA-sequencing datasets, and a large genealogical kinship dataset derived from BALSAC-CARTaGENE. Compared with state-of-the-art dimensionality-reduction methods, CosMAP produces more coherent visual representations, improves neighbourhood preservation, and provides clearer global organization of digit classes, biological cell populations, and regional genealogical patterns. These results indicate that CosMAP offers a robust framework for exploratory analysis of complex, sparse, high-dimensional data. The implementation is publicly available at https://github.com/FenosoaRandrianjatovo/CosMAP-dr.

Tree Search-based test-time scaling of LLMs is a powerful tool for automated scientific coding. However, pure Tree Search sometimes struggles with systematic exploration, becoming trapped in local optima, or unproductive loops, especially in the vast search space of scientific methods. To address this limitation, we propose Idea Search, a framework that systematically integrates a dynamic "Idea Bank" into Tree Search. Idea Search involves three steps: (1) decomposing existing methods into atomic ideas, (2) sampling from this bank of ideas to guide branches of code mutations, and (3) dynamically updating the bank with new ideas discovered through execution. On single-cell RNA-sequencing (scRNA-seq) batch integration, Idea Search reliably breaks the plateau of a strong pure Tree Search baseline, improving the mean score from 0.678 to 0.697 and reaching a best score of 0.728. We then characterize which design choices drive these gains: bank augmentation helps bandit sampling but not random sampling, "Exploratory" prompting that prioritizes new ideas surfaces the rare best-performing solutions, while increasing sampling-level exploration is counterproductive.

Spatial transcriptomics can resolve gene expression at single-cell resolution, but it is costly, limited to targeted panels of a few hundred to a few thousand genes, and applicable to only a small number of samples. H&E imaging, by contrast, is cheap and collected routinely at scale. This makes predicting single-cell expression directly from morphology a practical way to bring molecular analysis to large tissue archives. We therefore present VOICE, a multimodal foundation model that predicts single-cell gene expression from H&E images using paired Xenium data. VOICE first aligns cell centered H&E morphology from a pathology foundation model with single-cell expression embeddings from a transcriptome foundation model, trained using contrastive learning over 23 million cells. Next it predicts expression through two branches. One branch directly regresses expression from morphology. The other branch retrieves measured expression from similar reference cells, recovering genes that do not have morphological signal. Because genes vary in morphological predictability, VOICE fuses the two branches with a per-gene weight. After training, VOICE generalizes to heldout patients, slides, and partially overlapping gene panels from Xenium, and it consistently outperforms prior single-cell expression prediction methods on seven metrics.

5-aminolevulinic acid (5-ALA) promotes fluorescence-based resection of glioblastoma via protoporphyrin IX (PpIX); however, there is considerable heterogeneity in fluorescence intensity, limiting margin distinction. Cell-line studies show that stimulation of the MAPK pathway results in decreased PpIX levels through increased elimination by ABCB1, an efflux transporter, and ferrochelatase (FECH), the enzyme that converts PpIX to heme. However, the mechanism has yet to be studied in human tissue. Using two independent, purity-adjusted sets of primary IDH-wildtype glioblastoma specimens (TCGA-GBM, n=140, discovery; CGGA, n=87, validation), no reproducible correlation between MAPK signaling and its proposed downstream effectors (ABCB1 and FECH) could be detected. Instead, MAPK activity displayed a reproducibly negative relationship with PPOX, the enzyme that converts protoporphyrinogen IX to protoporphyrin IX: TCGA-GBM (n=133 with purity estimates; rho = -0.40, 95% bootstrap CI [-0.54, -0.25], adjusted p < 0.001), CGGA (rho = -0.28, 95% bootstrap CI [-0.47, -0.07], adjusted p = 0.0342). The finding was robust to permutation testing and to alternative approaches to purity adjustment. Here we report, for the first time, replicated human-tissue evidence for this association. If this correlation reflects causality, MAPK regulation of PpIX would act at the point of synthesis rather than at the clearance steps implied by previous cell-culture studies. We observed differential activation of the MAPK pathway with increased activation in the classical subtype and decreased activation in the proneural subtype. There was no correlation between MAPK pathway activation and overall survival (log-rank p = 0.73). The association between PPOX expression and MAPK pathway activation is biological rather than prognostic, identifying the MAPK pathway as a candidate handle for increasing 5-ALA fluorescence.

Classical likelihood-ratio tests and $\Delta$AIC exacerbate the statistical significance crisis by scaling with sample size, often flagging negligible improvements as highly significant. While causal estimands like the average treatment effect (ATE) quantify practical magnitude, their reliance on the expectation operator ties them to the data's original coordinate scale. Furthermore, existing pseudo-$R^2$ metrics are inadequate: variance-based measures ignore higher-order distributional changes, and current formulations lack invariance to monotone transformations. We resolve these limitations by introducing Entropic Variance (EV) as a rigorous, scale-independent generalization of error variance in ordinary least squares. We define the population EV-based parameter, $\rho^2_V$, which projects unbounded cross-entropy onto a standardized $[0,1]$ scale, and establish that the EV-based $F_\text{V}$ statistic asymptotically follows an $F$-distribution. Building on these distributional properties, we propose two estimators: the empirical population $R^2_{\text{SV}}$ and the out-of-sample predictive $R^2_{\text{SVP}}$. Both are derived by exponentiating per-observation cross-entropy and incorporate a degrees-of-freedom correction for training optimism. Leveraging the $F_\text{V}$-distribution, we derive refined $p$-values and confidence intervals for $\rho^2_V$ without requiring intractable Fisher information matrices. Simulation studies and a Parkinson's disease microbiome application demonstrate the superiority of variable selection via these EV-$R^2$ metrics. Notably, evaluating the $R^2_{\text{SVP}}$ of a LASSO path via data-splitting reduced false discovery rates from 80% to 6% in simulations while fully preserving signal recall.

Epitopes determine where antibodies bind antigens and shape downstream therapeutic properties such as functional blockade and escape resistance, making epitope understanding central to antibody drug discovery. Although large language models (LLMs) have shown strong biomedical reasoning ability, it remains unclear whether they can infer epitope information directly from antigen and antibody sequences. Existing epitope resources typically focus on isolated prediction tasks or rely on specialized structural settings, while general protein benchmarks do not evaluate epitope-centered decisions across the antibody development workflow. To address this gap, we introduce EpiBench, a closed-book, sequence-based, and automatically scorable benchmark for evaluating epitope reasoning in LLMs. EpiBench contains 1,609 curated samples grounded in structural antibody–antigen contacts, curated functional B-cell assays, and deep mutational scanning escape measurements. It covers five connected tasks: targetable region discovery, antibody-conditioned epitope identification, epitope binning, functional epitope assessment, and antibody escape assessment, with controlled sampling to reduce shortcut-based evaluation artifacts. We evaluate nine general-purpose LLMs and analyze their behavior through task-specific baselines, antigen length stratification, explicit-reasoning comparison, and failure-mode inspection. The results show that current LLMs capture partial epitope-related signals but remain limited in antibody-specific sequence grounding, long-context residue localization, and biologically grounded reasoning. Therefore, EpiBench provides a diagnostic testbed for measuring and improving sequence-aware biomedical LLMs toward reliable LLM-assisted antibody discovery.

Genomic foundation models are increasingly reused as frozen feature extractors for downstream sequence prediction, offering a compute-efficient alternative to full fine-tuning. However, it remains unclear when biological information encoded by these models is accessible without task-specific adaptation. We present a representation-accessibility analysis of frozen genomic language models across regulatory, epigenetic, promoter, splice-site, and variant-effect prediction tasks. We evaluate DNABERT-2, Nucleotide Transformer, HyenaDNA, GENERATOR-v2, and Omni-DNA under unified frozen-probing protocols, while separating diagnostic readout analyses from validation-selected checks. Our results reveal a consistent task-dependent pattern: frozen probes recover 95-100 % of fine-tuned performance on promoter tasks, but average splice-site recovery drops to 60-88 %. Frozen embeddings are also competitive on broad Genomic Benchmark tasks such as coding-region and species-discrimination classification, but show larger gaps on some regulatory and OCR tasks. Layer-wise probing, in-silico mutagenesis, variant-effect prediction, and embedding geometry show that local biological signal is partially present in frozen representations, but is not always accessible through final pooled embeddings.

A natural language interface can be used to make cancer genomics databases easier to use, but even if a question is perfectly fluent, its scientific meaning can be ambiguous. We propose CLARA, a framework that represents a question as a typed scientific query specification, considers a few possible interpretations, executes them, and asks for clarification when the estimates diverge. CLARA was assessed on mutation-prevalence contrasts among eight TCGA PanCancer Atlas cohorts and a 30-gene panel. This benchmark consisted of 330 unique executable contrasts varying in mutation scope, assay denominator, and sample context; 115 contrasts were result-sensitive and 215 were result-stable, per the preregistered definition of relative divergence greater than 0.10 or absolute divergence greater than 5 percentage points. An independently implemented pandas execution engine perfectly replicated all 660 results from the SQLite engine. In a separate 120-question LLM-generated, manually vetted language stress test, CLARA recognized all 60 result-sensitive contrasts and needlessly clarified 13 of 60 stable contrasts (accuracy 89.2%, sensitivity/recall 100%, specificity 78.3%). Standalone machine learning had superior overall accuracy (97.5%) but missed one critical contrast. This demonstrates that downstream execution can distinguish consequential from inconsequential ambiguity and reveal an explicit trade-off between safety and burden.

We study a self-supervised generation task for single-cell gene expression vectors: given a set of vectors from a cell type, we aim to generate additional gene expression vectors of that cell type. For this task we characterize both the biological fidelity of the generated gene expression vectors and the scaling behavior of the pretraining loss. The model is a causal transformer paired with a learned quantized VAE tokenizer, trained with a cross-entropy loss. To evaluate the model, we condition it on held-out gene expression vectors of a cell type and generate vectors of gene expression, comparing the resulting distribution over gene expression vectors to the ground truth distribution of that cell type. We study the scaling properties of the proposed architecture by varying the number of trained parameters and the amount of training data. To our knowledge, we find the first jointly-fit two-exponent scaling law and compute-optimal frontier for a single-cell foundation model. Finally, we discuss how this pretrained model could be finetuned for perturbation response prediction.