---
title: "bionexus-reliability"
description: "Scientific Reliability Layer for AI-Assisted Biology. Know what your evidence actually warrants: pre-submission shadow review of multi-donor single-cell differential expression, evidence gap location, and human scientifi"
canonical: https://agentpluginsdirectory.com/plugins/bionexus-reliability
last-updated: 2026-10-04
---

# bionexus-reliability
Scientific Reliability Layer for AI-Assisted Biology. Know what your evidence actually warrants: pre-submission shadow review of multi-donor single-cell differential expression, evidence gap location, and human scientific review support.
- Slug: bionexus-reliability
- Publisher: BioNexus Team
- Repository: https://github.com/HERRY423/BioNexus
- Manifest: plugin.json
- Version: 1.0.0-rc.8
- License: Apache-2.0
- Category (editorial): research
- Skills: 10 (external-evidence-audit, instrument-data-to-allotrope, nextflow-development, provenance-and-audit, scientific-problem-selection, scvi-tools, single-cell-de-audit, single-cell-rna-qc, spatial-transcriptomics, start)
- MCP servers: 1 (bionexus-local-mcp)
- Stars: 31
- Repository created: 2026-08-14
- Repository last pushed: 2026-09-25
- Publisher type: User
- Listing: https://agentpluginsdirectory.com/plugins/bionexus-reliability
- Schema: https://agent-plugins.org/schemas/1.0.0/plugin.schema.json

## What bionexus-reliability does, in the publisher's words

> Know what your evidence actually warrants.

BioNexus is a warrant-first scientific reliability layer for AI-assisted bioinformatics. It audits analytical assumptions, calibrates evidence strength, caps unsupported claims, verifies execution provenance, and abstains when evidence is insufficient.

Not another AI scientist or workflow platform. BioNexus sits between AI-generated analyses and scientific claims.

- Design Identifiability: A paired or isogenic design (e.g., treated vs control within the same 2 donors or cell lines) can legitimately identify strong candidate signals with low within-donor dispersion.
- Effect-Size Regime: A deterministic monogenic knockout ($\text{Log2FC} > 6, \text{FDR} < 10^{-15}$) requires far less replication to rule out technical noise than subtle polygenic shifts ($\text{Log2FC} = 0.3$).
- Confounding vs Power: Having $N=10$ donors with unmodeled batch confounding or extreme uncorrected dispersion does not justify population claims.

From the project README, punctuation lightly normalized. Full text: https://raw.githubusercontent.com/HERRY423/BioNexus/HEAD/README.md

## Skills

- external-evidence-audit: Audit completed results from literature, database, analysis, sequence, structure, or slide capabilities, then optionally assemble an explicitly adjudicated multi-source claim assessment. Passive only; never select tools, infer evidence relationships, or make the final scientific decision.
- instrument-data-to-allotrope: Convert laboratory instrument output files (PDF, CSV, Excel, TXT) to Allotrope Simple Model (ASM) JSON format or flattened 2D CSV using native allotropy or declarative YAML mapping rules. Use this skill when scientists need to standardize instrument data for LIMS systems, data lakes, or downstream…
- nextflow-development: Prepare nf-core samplesheets, cluster nextflow.config, and write launch artifacts. nfcore_launch.py only wraps nf-core/rnaseq and nf-core/scrnaseq with optional nextflow -preview. Other pipelines still use generate_samplesheet.py plus a hand-written nextflow run. Does not reimplement the pipelines.
- provenance-and-audit: SHA-256 hashes, environment snapshot, and activity-aware Methods text. Use to attach a reproducibility sidecar to an analysis. Do not use for 21 CFR Part 11, GxP, ALCOA+, or CLIA audit claims.
- scientific-problem-selection: This skill should be used when scientists need help with research problem selection, project ideation, troubleshooting stuck projects, or strategic scientific decisions. Use this skill when users ask to pitch a new research idea, work through a project problem, evaluate project risks, plan research…
- scvi-tools: Train official scvi-tools models (scVI/scANVI/totalVI/PeakVI/MultiVI/veloVI) on raw counts after the scRNA gold chain. Use for probabilistic batch integration or those named models. Does not run DestVI/Cell2location. Do not log-normalize before setup_anndata.
- single-cell-de-audit: Evidence audit for multi-donor single-cell differential expression before lab meetings, manuscript submission, or data sharing. Users keep Scanpy, Seurat, or existing workflows; BioNexus audits pseudoreplication, donor replicates, donor cell imbalance, batch confounding, raw count layers, and FDR c…
- single-cell-rna-qc: scverse scRNA gold chain on.h5ad/.h5: inspect, convert, MAD QC, optional scanpy.pp.scrublet, preprocess, Harmony/ComBat, PCA-UMAP-Leiden, Wilcoxon markers, subset, pseudobulk, pydeseq2, stable plots. Use when the user has scRNA-seq counts. Does not assign cell-type labels. rank_genes_groups is no…
- spatial-transcriptomics: squidpy spatial gold chain on SpatialData.zarr or AnnData.h5ad with obsm['spatial']. Use when the user has Visium/Slide-seq/generic spots or cells with coordinates. Builds a knn spatial graph, Moran SVGs, and spatial_scatter plots. Multi-table SpatialData requires --table. Does not run Cell2locat…
- start: Orient a session on this plugin. Use first. Run scripts/doctor.py, then route only to core gold-chain skills unless the user names a heuristic job. Do not assign cell-type labels. Do not use this skill to run analyses.

Descriptions come from the frontmatter of each SKILL.md, punctuation lightly normalized.

## MCP servers

- bionexus-local-mcp: transport: stdio; command: python ${PLUGIN_ROOT}/scripts/local_mcp_server.py --audit-log ${PLUGIN_ROOT}/.bionexus-audit/mcp-host-audit.jsonl

Read from the plugin's own mcp.json. Environment variable names only, never values.