LATEST IN SCIENCE

Abstract: Selectively eliminating cancer cells based on their oncogenic drivers holds great promise but remains challenging for targeted cancer therapy. Here, we present ORCA (Oncogenic RNA-triggered Cancer Ablation), a programmable and modular platform that couples RADAR-based detection of oncogenic RNA with conditional expression of therapeutic payloads in cancer cells. We identify RADAR sensors targeting the transcripts of endogenously amplified oncogenes, including MYC, FGFR2 and wild-type KRAS in colorectal and gastric cancer cells. Focusing on colorectal cancer (CRC) driven by extrachromosomal DNA-amplified MYC, we demonstrate targeted ablation of high-MYC cancer cells using two orthogonal strategies: anti-CD19 CAR-T therapy and enzyme-prodrug therapy. In vivo delivery of ORCA using adeno-associated virus (AAV) suppresses tumor growth in a subcutaneous xenograft model and prolongs survival in a CRC liver metastasis model. ORCA establishes oncogenic RNAs as actionable inputs for targeted cancer therapy via programmable RNA sensing, thus offering a strategy to target tumors through the transcripts of otherwise difficult-to-drug oncogenes such as MYC.

Abstract:

Abstract: -Keto-{beta}-amide warheads are protease-inhibiting pharmacophores in peptidomimetic antiviral drugs and drug leads, but their diversity-oriented synthesis for library-based drug discovery can be challenging. The same moiety is installed enzymatically in gene-encoded peptides by tyramine spliceases from spliceotide natural products. However, their substrate scope remains poorly understood. We mined bacterial spliceotide pathways for multi-precursor spliceases and identified HypX from Hyella patelloides that modified 5 of 7 native precursors in E. coli. Single-point mutagenesis showed promiscuity for HypX and the previously identified PlpX but different substrate scopes, suggesting that splicease mining can provide enzymes for focused libraries of various substrate types. To evaluate multipoint variants without exhaustive screening, we combined AlphaFold3 models with Rosetta calculations to rank millions of variants. Top-scoring peptides were spliced more efficiently than single-point mutants. Structural modeling rationalized position-specific tolerance, including a cofactor constraint. This work establishes a mining- and structure-guided strategy to streamline library-based splicease applications.

Abstract: Phylogenetic analyses of molecular evolution often require repeated manual preparation to reconcile taxa, annotate trees, execute models, and summarize results. These steps become increasingly time-consuming and difficult to standardize as the number of taxa, genes, and analytical hypotheses increases. We present HYphlow, an open-source graphical workflow developed in Python that integrates data preparation, trait-based foreground branch assignment, HyPhy execution, and structured result extraction within a single interface. HYphlow reconciles taxon labels across alignments, trees, and trait data; reconstructs ancestral states using parsimony and maximum likelihood; generates alternative foreground annotations; executes HyPhy analyses locally or through portable scripts for high-performance computing; and records analytical decisions and settings throughout the workflow. We evaluated HYphlow using simulations and two published datasets comprising cryptochrome analyses in pigeons and doves and visual opsin analyses in skinks. A name-similarity threshold of 80 reliably recovered simulated spelling variants, while consensus foreground assignment increased precision relative to parsimony or maximum likelihood alone (0.928 vs 0.891 and 0.907, respectively), with a corresponding reduction in recall (0.875 vs 0.921 and 0.918). In the empirical datasets, HYphlow reconciled taxon labels, reconstructed or reproduced foreground hypotheses derived using different approaches, and generated selection results consistent with the principal conclusions of the original studies. HYphlow therefore provides a reproducible framework for carrying comparative molecular evolutionary analyses from independently prepared sequence, phylogenetic, and trait data through branch assignment, model execution, and results summary while retaining researcher control over analytical decisions. HYphlow is freely available under the MIT licence from GitHub at: https://github.com/Schott-Lab/HYphlow.

Abstract: Mechanistic inference of gene regulatory networks from temporal transcriptomic data can become computationally prohibitive at transcriptome scale, making gene selection an important determinant of which regulatory relationships can be recovered. However, gene selection for GRN inference commonly relies on static expression or co-expression structure, rather than prioritizing dynamically informative genes whose temporal patterns capture how transcriptional programs change and relate to one another. We introduce GSET (Gene Signals from Embedding Trajectories), an upstream retrieval framework that extracts dynamic information from temporal gene embeddings using two complementary signals: an optimal transport residual, which captures gene-specific deviations from global embedding movement, and a KL divergence, which detects changes in local co-expression neighborhoods. Temporal gene-embedding similarity identifies static transcriptional programs, whereas vectorial GSET signals represent coordinated temporal changes among genes, complementing this static structure. Lagged comparisons of these signals further prioritize temporal relationships between programs and candidate bridge genes associated with their transitions. Integrated with CardamomOT, GSET combines static program cores with dynamically retrieved bridge genes into compact, program-informed gene sets for mechanistic GRN inference. In synthetic benchmarks, at equal gene budget these sets recover more of the regulatory links between programs than static selections of the same size, and inference on the edges within the set is comparable to that of the full-gene model on the same gene pairs. On a time course of mouse fibroblasts reprogrammed into induced pluripotent stem cells, GSET traces the transition from the fibroblast program to the cell-cycle program. By providing a dynamic retrieval layer upstream of GRN inference, GSET nominates bridge genes as candidate regulators of cell-state transitions, which, once refined by a mechanistic model such as CardamomOT and confirmed by perturbation experiments, could open the way to controlling these transitions.

Abstract: Frontier large language models (LLMs) trained with extended reasoning, such as DeepSeek-R1 and OpenAI o1, have established state-of-the-art benchmarks in formal logic, competitive programming, and mathematics. In interactive, empirical scientific sandboxes, however, their efficacy remains counter-intuitive. In the Bioinformatics Benchmark (BixBench), extended reasoning models paradoxically degraded in multi-turn environments, leading benchmark authors to exclude them from primary agentic evaluations. In this paper, we conduct the first systematic forensic audit of the internal cognitive monologue ( ... ) of open-weight reasoning models (DeepSeek-R1-Distill-Qwen-14B) and matched instruction-tuned controls (Qwen2.5-Coder-14B) deployed inside containerized computational biology sandboxes on Nvidia H100 GPUs. Our forensic analysis uncovers two distinct systemic failure modes: (1) The PubMed Prior Trap, in which reasoning models engage in epistemic rationalization within their cognitive stream, falsely convincing themselves that empirical datasets are missing and substituting empirical computation with speculative literature recall (e.g., estimating adjusted p-values based on "typical experimental outcomes"); and (2) Tag Language Nesting Collisions, in which models conflate agent-level XML action tags with Python syntax, producing terminal interpreter syntax errors. To resolve this paradox, we introduce Dual Track Decoupled Reasoning (DTDR), an agent scaffolding architecture that evicts internal cognitive monologues from multi-turn history, injects strict empirical grounding guardrails, and dynamically intercepts runtime tool invocations. On a stratified diagnostic benchmark spanning single-cell RNA-seq, bulk differential expression (DESeq2), and genomic variant prioritization (n=22 per arm), DTDR rescues the 14B reasoning model from 0.0% to 9.09% accuracy (2/22), doubling the performance of the matched non-reasoning coder control (4.55%, 1/22) while slashing code execution errors by 81% (13.6% vs. 72.7%) and reducing average steps to solution by 66% (1.77 vs. 5.18 steps). All replication pipelines, prompts, and trajectory manifests are openly released.

Abstract:

Abstract: Identifying and quantifying molecular components of biological fluids offers great potential for the diagnosis of disease, detecting environmental contaminants, and process monitoring. The use of simple 1D 1H NMR spectra to accomplish this task offers advantages in terms of broad applicability and quantitative output. However, it also encounters obstacles in extracting individual components from complex superimposed spectra and in compound identification through comparison with limited database entries. Here, we present a new protocol that includes automatically separating component spectra using singular value decomposition (SVD) and matching them to database entries that carry both chemical shift and scalar coupling information. The procedure is illustrated by application to a fungal extract from Neurospora crassa that is chromatographically fractionated to form a two-dimensional spectral fraction library. Results are compared to a previous analysis that used a semi-manual curation of separated spectra followed by chemical shift matching.

Abstract:

Abstract:

Abstract: The clinical management of Barrett's esophagus (BE) and the risk of esophageal adenocarcinoma (EAC) is dependent on the presence or absence of dysplasia. The histopathological diagnosis of dysplasia is unreliable however, and would benefit from additional diagnostic criteria. Here, we obtained EM-seq whole-genome methylation profiles from 86 samples across the BE/EAC disease spectrum, notably supplementing the scant existing methylation evidence for low-grade and high-grade dysplasia. While the bulk of methylation changes occur in the transition from normal squamous epithelium to intestinal metaplasia (non-dysplastic Barrett's esophagus), we identified a smaller set of changes that mark the onset of dysplasia and progression to EAC. Methylation changes were broadly independent of copy number variation. They were tightly anticorrelated with RNA-seq expression for some genes. Deconvolution of the bulk methylation signatures into constituent cell types revealed four sample groupings enriched with a particular cell type (squamous, columnar, lymphoid and erythroid cells), somewhat reflecting existing clinical classifications. We outline potential markers that could improve the diagnosis of metaplasia and dysplasia in EAC.

Abstract: Gene expression patterns within the same tissue region can exhibit similar correlation or enrichment, but still differ in their organization within regions and at their boundaries. We present TileScope, a framework for analysing genes and transcriptional programs within predefined tissue regions, their components and their contacts with neighbouring regions. TileScope combines regional enrichment, within-region correlation, cross-correlation over distance and expression differences near specific boundaries. Repeated spatial partitions and consensus aggregation show how these measurements depend on the grouping of spots or cells. Across 30 controlled constructions derived from glioblastoma and hepatocellular carcinoma data representing ten spatial expression scenarios, combined measurements separated arrangements that individual statistics left ambiguous, including coincident patches, gradients and expression localised at region boundaries. In 30 human glioblastoma sections, TileScope distinguished shared regional enrichment from internal covariation and identified hypoxia-associated expression localised at region boundary despite negligible within-region correlation. In newly acquired Visium profiles of control and myeloid-targeted MCT1-knockout (KO) orthotopic GL261 mouse gliomas, proliferation and myeloid programs were enriched in tumour peripheries but showed opposite within-periphery-region correlations. Proliferation was higher near the tumour-facing boundary in both samples, whereas myeloid expression was higher near the tumour-facing boundary in the control and farther within periphery in the KO condition. Xenium applications on glioblastoma and breast cancer samples demonstrated feasibility with irregularly spaced individual cells and different regional annotations. TileScope provides spatially explicit descriptions of gene and program relationships, linking their organization within tissue compartments to their sensitivity to aggregation.

Abstract: Cisplatin-based neoadjuvant chemotherapy remains an important treatment for advanced bladder cancer (BLCA), yet approximately half of patients derive limited benefit because of chemoresistance. Although DNA damage repair (DDR) is a key determinant of this resistance, how DDR heterogeneity contributes to chemoresistance remains unclear. Single-cell transcriptomic analysis revealed DDR heterogeneity among BLCA epithelial cells, with epithelial cells exhibiting higher DDR activity showing greater malignant potential. Bulk transcriptomic and proteomic analyses identified two molecular subtypes of BLCA. Compared with the DDR-Base subtype, the DDR-Active subtype exhibited enhanced DDR function and worse clinical outcomes. Integrative machine-learning modeling and SHapley Additive exPlanations (SHAP) prioritized GALNT6 as a glycosylation-related regulator associated with the DDR-Active phenotype. GALNT6 was highly expressed in the DDR-Active subtype and progressively increased along the epithelial-cell trajectory toward the DDR-enhanced malignant state. Functionally, GALNT6 knockdown impaired DNA damage repair efficiency, thereby overcoming cisplatin resistance in BLCA. Mechanistically, GALNT6 stabilized SMAD2 through O-glycosylation, after which SMAD2 transcriptionally activated CDC7 to enhance repair capacity against exogenous DNA damage. Moreover, GALNT6 and SMAD2 co-expression was a hallmark of the DDR-Active subtype and was associated with advanced tumor stage. This study defines clinically relevant DDR subtypes in BLCA and identifies the GALNT6/SMAD2/CDC7 axis as a driver of cisplatin resistance. These findings provide a biological framework for understanding heterogeneity in cisplatin response and for future treatment stratification.