岐黄QH-Omic
Computing for life.
An agentic expert system for non-model organism research that connects research goals, real computation and traceable evidence.
Researchers state their goals in natural language in a browser workbench. Agents plan, call tools, coordinate tasks and explain results; bioinformatics software does the actual computation; and the system keeps the evidence linking inputs, methods, execution records and results.
Try it: this is the QH-Omic Workbench.
Not a screenshot: a Workbench you can operate. Send a message in the Channel and watch the Orchestration Agent split the analysis, assemble a team, run preflight and stop when it needs your authorization. Switch to Plugins to browse the capability catalogue, or to Tasks to follow coordination.
Acorus flower–leaf transcriptome# flower-vs-leafChannel ready · Orchestration Agent online
Agents in this Channel
- Orchestration AgentIdle
Run evidence
Once preflight passes and you authorize the run, execution status, limitations and artifacts appear here.
Interactive demo: whatever you type, the Channel replies with the same preset script. No model is called and nothing is computed. Analysis figures come from the computational verification records of an Acorus transcriptome; the run, its timing and the people are illustrative. The plugin catalogue is the snapshot used for the demo; actual capabilities depend on the environment.
One research goal, five reviewable steps.
Confirm the research context
Check species, samples, treatments, biological replicates, batches and comparisons; when something is missing, list it explicitly.
Form an analysis plan
Pick capabilities from installed, permitted plugins, declaring inputs, methods, reference resources, outputs and dependencies.
Preflight before running
Check data, software, reference versions, resources and experimental design; missing essentials stay as recorded blockers, never guessed by a model.
Run real computation
Run the tools in the configured local, remote or Slurm environment, recording execution and resource facts.
Read back and interpret
Agents reread the saved results and organize an interpretation with literature and review records, showing conflicts, uncertainty and next validation steps.
What it can do today
The most complete line today is Bulk RNA-seq and non-model organism transcriptome analysis. What actually runs depends on the chosen capabilities, installed tools, databases, references and environment.
Bulk RNA-seq
Read QC, Salmon quantification, sample quality checks and DESeq2 contrasts.
Non-model references & annotation
Reference-guided transcriptome assembly, EviAnn structural annotation, TransDecoder / Trinotate / eggNOG functional annotation.
RNA events & non-coding RNA
rMATS / ASGAL splicing, APA, circRNA, lncRNA, miRNA target prediction, CPC2 and Rfam screening.
Enrichment & transcriptome synthesis
GO / KEGG enrichment, multi-branch synthesis and evidence-conflict review, keeping test backgrounds and multiple-testing correction.
Single-cell RNA-seq
10x matrix intake and QC, normalization, PCA, kNN, Leiden, marker candidates and pseudobulk differential expression.
Pooled CRISPR screens
MAGeCK-based input checks, screen design, execution and effect output.
Literature & public data
PubMed / PMC search, full-text follow-up, inclusion review, siRNA evidence extraction and public RNA-seq dataset discovery.
Data identity & research organization
Paper and accession checks, supplementary retrieval, sequence and value comparison, and a Research Program evidence index.
Execution finished is not a conclusion proven.
The system judges every piece of research work at three levels, and keeps any unmet limit visible in the results.
- Execution completeDid the tools actually run, and are outputs and execution records complete?
- Agentic completeIs there a full evidence chain of planning, tool choice, task binding, result read-back and interpretation?
- Scientific eligibilityDo design, statistics, identity mapping, evidence independence and review support this specific claim?
Case: computational verification of an Acorus transcriptome
These are computational verification records, not wet-lab confirmation; the original records also keep the questions that remain open.
| Analysis | Recorded computational result | What it can support |
|---|---|---|
| Pathway checks after differential expression | 28,914 gene rows as one test background, compared across several backgrounds; leaf photosynthesis / antenna and flower phenylpropanoid enrichment keep support | Tissue-related expression and function signals; batch, donor and annotation bias still apply |
| Correcting an enrichment claim | Flavonoid and sesquiterpene pathways did not pass multiple-testing correction | The 'significantly enriched pathway' wording was withdrawn; evidence for specific candidates was kept |
| circRNA structure checks | 4,535 distinct junction IDs; of 97 exon-labelled candidates screened further, 8 have a unique fully annotated exon path | Structural support for some candidates; the rest stay undetermined |
| lncRNA candidate screening | CPC2, Rfam and reference CDS checks added for 3,862 candidates, giving 2,739 conservative priority candidates | A follow-up validation set, with coding-potential and annotation conflicts preserved |
Who it is for
Non-model organism researchers
Organize transcriptome studies with incomplete references and many analysis branches, keeping reference choices and interpretive limits visible.
Bioinformaticians and labs
Connect data, tools, models and compute around one research task, with run dossiers that are easy to review and hand over.
Research-agent and method developers
Add new analysis capabilities through plugin contracts, reusing task, permission, execution and evidence infrastructure.
The QH-Omic icons
A duotone set drawn for QH-Omic: 48 px grid, 2 px line, lime accent. Select any icon to download its SVG.
