bioRxiv Science⌕ Search
Home/Archive/bioRxiv

bioRxiv · 10.64898/2026.10.16.759698

VIGAT: VLM-guided gene activity quantification from single-cell ATAC-seq

Ji, B.·Ji, Z.
View original source ↗Export metadata ↓

Abstract

Summary: Gene activity quantification from single-cell ATAC-seq is challenging because genomic regions informative of cell identity can vary across genes and tissues, while broadly accessible regions may provide limited cell-type-discriminative information. We developed VIGAT, a vision-language model (VLM)-guided tool that automates visual interpretation of ATAC coverage plots for adaptive gene activity quantification. VIGAT jointly examines accessibility patterns across multiple marker genes, integrating genomic annotations and biological context to select gene-specific regions enriched for cluster-discriminative signals. Selected regions are used to generate a gene-by-cell activity matrix. Across 196 paired single-cell RNA-seq and ATAC-seq datasets, VIGAT outperformed five existing methods in recovering marker-associated accessibility enriched in the corresponding cell populations, demonstrating the utility of VLM-guided visual reasoning for scalable gene activity quantification. Availability and implementation: VIGAT is implemented as an open-source R package and is freely available at https://github.com/beijie02/VIGAT.

Explore related subjects

bioinformaticsPreprint

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Ji, B., Ji, Z.. 2026-11-03. VIGAT: VLM-guided gene activity quantification from single-cell ATAC-seq. https://doi.org/10.64898/2026.10.16.759698

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2026

Improving real-time search through deep learning-enhanced fragment ion indexing

The fragment ion indexed database searching approach enables highly efficient tandem mass spectrometry database searches, including real-time search workflows. However, the index can require tens of gigabytes of memory, often exceeding the free memory available on instrument acquisition computers running the search. This restricts the size of the searchable database, the variable modification space, and the number of searches that can run concurrently. To address this problem, we developed an index-pruning pipeline within the Comet database search tool using Carafe, a deep learning tool that predicts fragment ion intensities based on peptide sequences. In our approach, only fragment ion positions that are predicted to be observable are written in the index, eliminating entries that are unlikely to contribute to spectrum detection. We evaluated this approach on two search spaces in three index configurations to demonstrate that masking significantly reduced the number of fragment ion index entries and, consequently, peak search memory use. Masked searches were also faster than unmasked searches and identified a comparable number of peptide-spectrum matches. Because all prediction and masking steps are performed before searching, their cost is amortized across repeated, future searches of the same database and modification configuration, introducing no per-spectrum search overhead. The complete implementation is open source with an Apache license.

bioinformatics↗
bioRxiv2026

FastaGuard: FASTA preflight QC for modern bioinformatics pipelines

Summary: FastaGuard is a streaming, explainable FASTA preflight QC tool that validates assembly FASTA files, detects structural and composition red flags, and emits a versioned multi-format report with an explicit pipeline gate before expensive downstream analysis. We evaluated v1.1.0 in four analyses whose design was fixed before data collection. On 700 public NCBI assemblies the pipeline gate stopped 19.7% (1.9% weighted to the population), always on its high-N rule and mostly in scaffold-level genomes; the NCBI submission gate stopped 2.0%, and an independent recount confirmed every stop. Among 1,040 faults injected into real assemblies, the NCBI gate flagged 520 of 520 in 13 error classes (table2asn: 280 at error level) and blocked none of 120 benign changes. Across seven downstream tools, 46 of 182 defect cells failed and 12 altered the records without a warning, all in classes the pipeline gate blocks. Stopping a mismatched reference at the gate avoided about 54-62 s of mapping and sorting at a cost of 0.05-0.06 s. Availability and Implementation: Source code and documentation are freely available under the MIT license at https://github.com/ehsanestaji/FastaGuard (release tag v1.1.0: https://github.com/ehsanestaji/FastaGuard/releases/tag/v1.1.0; archived at doi:10.5281/zenodo.23728753). Install via Bioconda (mamba install -c conda-forge -c bioconda fastaguard=1.1.0), BioContainers, or GitHub release binaries. Implemented in Rust; supported on Linux and macOS, including Bioconda builds for linux-64, linux-aarch64, osx-64, and osx-arm64. Contact: ehsan.estaji@umu.se; jianfeng.mao@umu.se Supplementary information: Supplementary Note (designs, all results, deviations from the pre-registered protocol); scripts, manifests and result tables at Zenodo, doi:10.5281/zenodo.23729530.

bioinformatics↗
bioRxiv2026

Follow-up availability and the robustness of immune transcriptomic recovery after myocardial infarction

Background: Longitudinal blood transcriptomes describe immune changes after myocardial infarction, but later assays do not necessarily represent the population sampled at admission. We examined whether recovery summaries depend on observable selection and assumptions about unavailable assays. Methods: We reanalyzed official processed data from GSE59867, linking 111 patients with ST-segment elevation myocardial infarction across admission, discharge, one month, and six months. Seven fixed Hallmark programs were scored on a common 17,590-gene vocabulary using a bounded within-sample rank statistic. The primary endpoint was the six-month inflammatory-response score. We separated paired-completer change from the observed follow-up mean minus the entire baseline mean; fitted cross-fitted availability and outcome models; and evaluated full-support completion bounds and prespecified missing-mean scenarios. Artificial deletion tested estimators in the 83 observed six-month participants. GSE123342 provided unlinked marginal comparisons and explicitly conditional completion scenarios. Results: GSE59867 contained 101, 95, and 83 follow-up assays. At six months, inflammatory-response change was -0.00893 among paired completers and -0.00999 for the marginal contrast (95% participant-bootstrap interval -0.01217 to -0.00783). Augmented inverse-probability weighting gave -0.01004 (-0.01556 to -0.00727). The full-support completion interval was [-0.16460,0.08765], but eliminating the negative change required an unavailable-assay mean 4.42 observed-follow-up standard deviations above the observed mean. The primary direction remained negative throughout the prespecified {+/-}2-SD scenarios; interferon and coagulation directions were less stable. Under artificial outcome-dependent deletion, all evaluated estimators retained bias. The unlinked external cohort had 65 admission, 64 day-30, and 37 year-one MI assays; the year-one inflammatory contrast was -0.00459, with a conditional tipping shift of 1.70 SD. Conclusions: The observed inflammatory-response rank-score decline was stable to baseline-score adjustment and the prespecified {+/-}2-observed-SD missing-mean scenarios, while unrestricted completion did not identify its direction. Robustness varied across programs and data settings. These findings concern relative transcriptomic scores and assay availability, not clinical recovery, survival, or intervention efficacy.

bioinformatics↗