Deep contrastive learning enables genome-wide virtual screening

Identifying drug candidates across the entire human proteome has been a long-standing computational challenge because traditional docking methods are too slow to operate at genomic scale. Researchers developed DrugCLIP, a contrastive learning model that learns shared representations of protein pockets and small molecules to enable extremely rapid virtual screening — up to ten million times faster than conventional docking — while matching or exceeding docking-based baselines on standard benchmarks. Wet-lab experiments confirmed the approach's practical value: DrugCLIP achieved a 15% hit rate against the norepinephrine transporter, with crystal structures solved for two discovered inhibitors, and a 17.5% hit rate against thyroid hormone receptor interactor 12 using only AlphaFold2-predicted structures rather than experimentally determined ones. To make these capabilities broadly accessible, the team released GenomeScreenDB, a public database containing precomputed screening results for roughly 10,000 human proteins against 500 million compounds, establishing a new paradigm for proteome-scale drug discovery.

Read the original article →

Mapping somatosensory afferent circuitry to bone identifies neurotrophic signals required for fracture healing

Bone fracture pain is transmitted by somatosensory neurons, which also appear necessary for bone regeneration. To map this relationship, researchers used single-cell transcriptomics to profile dorsal root ganglia neurons innervating mouse bones before and after fracture. CGRP+ and Aβ-Field LTMR neurons were the predominant bone-innervating populations. Gene expression in these sensory neurons shifted dynamically across the phases of bone repair, including upregulation of morphogens such as Tgfb1, Fgf9, and Shh. When innervation was experimentally removed, fracture healing was impaired, accompanied by deficient mesenchymal cell proliferation and bone differentiation. Building on these transcriptomic findings, the team identified FGF9 as a key driver of fracture repair. These results establish a direct link between sensory innervation and skeletal regeneration and point to FGF9 as a potential therapeutic target for promoting bone healing.

Read the original article →

Chinese Immune Multi-Omics Atlas

Researchers built the Chinese Immune Multi-Omics Atlas (CIMA) to understand how sex, age, and genetic variation shape molecular and cellular diversity in circulating immune cells. Using multi-omics profiling of more than 10 million peripheral blood immune cells from 428 Chinese adults, they mapped gene regulatory networks, identifying 237 enhancer-driven regulons, roughly 9,600 genes whose expression is influenced by local genetic variants (eGenes), and over 52,000 chromatin accessibility peaks associated with genetic variants—all resolved at the level of individual cell types. The study also uncovered overlapping associations between disease-linked genetic loci, expression quantitative trait loci, and chromatin accessibility QTLs. An additional cell language model, CIMA-CLM, was developed to predict chromatin accessibility and assess how noncoding genetic variants affect gene regulation. Together, these resources offer a detailed population-specific reference map that could improve understanding of genetic risk for immune-related diseases.

Read the original article →

IDH-mutant gliomas arise from glial progenitor cells harboring the initial driver mutation

Pinpointing which cell first acquires a cancer-driving mutation is critical for understanding how tumors develop and for designing better therapies. In IDH-mutant gliomas — the most frequent malignant brain tumor in young adults — this cell of origin has remained unclear. Researchers performed deep sequencing on 142 tissue samples from 70 patients, covering tumor, peritumoral cortex, subventricular zone, and blood. They detected low-level IDH mutations in the peritumoral cortex of roughly 38% of patients. By combining cell-type-specific mutation profiling, clonal evolution analysis, spatial transcriptomics from patient brain tissue, and a mouse model in which IDH mutations were introduced into oligodendrocyte progenitor cells, the team established that glial progenitor cells carrying the initial IDH mutation give rise to these gliomas. The findings clarify a long-standing question about glioma origin and may open new avenues for early detection and targeted intervention.

Read the original article →

Bark microbiota modulate climate-active gas fluxes in Australian forests

Microbes living on and within tree bark have received little scientific attention, leaving their ecological functions largely unknown. This study used gene-centric and genome-resolved metagenomics to characterize the microbial communities colonizing the bark of eight Australian tree species. The dominant bacteria are facultative anaerobes capable of cycling hydrogen and tolerating fluctuating oxygen and nutrient conditions. Notably, methane-consuming bacteria (methanotrophs) coexist alongside methane-producing archaea (methanogens) in bark. Laboratory microcosm experiments demonstrated that these microbes can aerobically consume methane, hydrogen, and carbon monoxide at concentrations naturally occurring in trees, while producing these gases under low-oxygen conditions. Field measurements confirmed that bark microbiota process climate-relevant gases at substantial rates directly within tree stems. The findings suggest that tree-dwelling microbial communities play a meaningful, previously overlooked role in regulating atmospheric concentrations of greenhouse and trace gases at a global scale.

Read the original article →