RPG
Research Process Graph Browser
Every Plant Cell paper, broken down into the questions it asks (Q), the methods used (M), and the findings produced (F) — fully cross-linked.
Expert similarity network
Every corresponding author in The Plant Cell is now a node in a similarity graph. Edges connect experts whose Q→M→F profiles overlap. Pick any researcher; their 20 most similar peers orbit them.
Who works on the same problems — even with different tools?
Who has the same wet-lab toolbox — even on different problems?
Who keeps producing the same kind of result?
We use weighted Jaccard similarity over each researcher's taxonomy profile (Q topics, M techniques, F categories). Hover any edge to see what's literally shared between two experts; hover a node to see that researcher's top Q/M/F items. Click a neighbor to walk the graph.
Open the network →Drag & drop a PDF here
or click to browse
PDF files accepted
No graphs yet. Upload a PDF to get started.
Loading samples...
The Plant Cell RPG Library
Browse -- pre-computed Research Process Graphs from The Plant Cell
Loading The Plant Cell library...
QMF Taxonomy Browser
Explore the hierarchical classification of -- research sentences across Questions, Methods, and Findings from The Plant Cell
Browse all tags — click any chip to search
Questions
Hypothesis classes derived from the question text. Click to find every question in that class.
Methods
304 methods grouped by their broader category. Click a method to see who uses it and which papers apply it.
Findings
Seven GO-like categories applied to findings (multi-tag). Click to search.
Loading taxonomy...
Expert Directory
Explore -- corresponding authors from The Plant Cell — browse experts by the methods they use, the questions they ask, and the findings they produce.
Loading researchers...
Enter a method — see who uses it and what they find
Search any method, technique, or approach. Our AI understands synonyms and related terms.
Select an example above or type your own query to get started.
Expert network — who works on related things?
Each node is a researcher. Edges connect experts who use related methods, ask similar questions, or produce similar findings. Pick what counts as “related” with the toggle below.
Question Topics (Q)
L2 topics from the LLM taxonomy: what this researcher's questions are actually about. Hover for the L1 context.
Techniques (M)
Actual technique names used, sized by how many of this researcher's papers used each.
Finding Tags (F)
7 GO-like categories on each finding (multi-tag).
Research Questions (Q)
Methods (M)
Findings (F)
Topic trajectory
How this researcher's question topics shifted year by year. Each stack is one paper's Q-L1 categories.
Related researchers
Other corresponding authors who use the most overlapping techniques. Click a name to open their profile.
Similar experts — network view
This researcher in the center, their top 20 similar peers around. Edges connect anyone whose Q/M/F profile overlaps. Click a neighbor to open their profile.
Browse Questions, Methods & Findings
Papers
Digital Professor
Enter a research question and get a data-driven research prospectus — a concise summary, recommended methods with linked findings, and follow-up questions with suggested methods — all grounded in 2,600+ Plant Cell papers.
What is RPG?
RPG = Research Process Graph. Every Plant Cell paper, broken down into the three things that make up a piece of plant-biology research: the Question the authors asked (Q), the Method they used (M), and the Finding they produced (F). Each node carries the verbatim sentence from the paper, plus an LLM taxonomy on top (technique tags on M, hypothesis-class tags on Q, GO-like tags on F). 2,600+ papers, ~150k nodes, all cross-linked through the Q→M→F DAG.
Three tools share this dataset. Each section below covers what it does, an example query, and how to read the result.
📚 Plant Cell Library
Browse every paper as a hierarchical Q | M | F table on top and a graph network below. Filter the catalog by year, by technique, or by question topic.
Try: open the catalog → pick technique = CRISPR/Cas9 → click any paper. The table shows every Q in the paper with its connected Ms and Fs, each row carrying the technique tags (green chips on M), GO-like tags (peach chips on F), and hypothesis-class tags (blue chips on Q).
How to read the paper view:
- Table (top): one row per Q, with rowspan grouping its Ms, and each M's row grouping its Fs. Click any cell to see the full node text.
- Network (bottom): the same graph as a Q→M→F DAG. Toggle layouts in the toolbar.
- Chips: M's green chips are technique tags from the upstream tagger (e.g.
RT-qPCR,confocal microscopy); F's peach chips are GO-like categories (Phenotype,Regulation); Q's blue chips are hypothesis classes (mechanism,function_test). - Related Papers at the bottom: weighted-Jaccard similarity over Q/M/F taxonomy categories. Slide the weights to re-rank.
🎓 Digital Professor
You type a research question; the tool returns a research prospectus grounded in real papers: ranked recommended techniques (each with the M-node text and the F it produced), follow-up questions with suggested methods, and an “Office Hours” chat that remembers the prospectus context.
Try: “How does my protein help plants cope with drought stress?”
How to read the result:
- Recommended techniques are ranked by paper count and Claude rerank.
Each card shows the LLM rationale, then concrete
M→Fexamples mined from real papers in the matched set. Click any paper title to open its RPG graph. - Follow-up Questions: each card is a real Q from a neighbouring paper, with its Suggested methods (the M-nodes wired to that Q in the source paper's DAG).
- Matched Questions: the seed Q-nodes that drove the prospectus. Click Run this question on any of them to re-run as a fresh prospectus.
- Office Hours chat: scoped to plant biology. Push back on a recommendation, tell it your constraints (“no confocal”, “6-week timeline”), or ask it to surface the Ms it has for a given question type. Refuses off-topic requests.
👨🏫 Find Experts
Search corresponding authors by the techniques, topics, and findings in their actual papers. Two paths: Browse All (list with technique-multiselect + name/year filters) and Find Experts (LLM-assisted method search).
Try: open Find Experts → type “chromatin immunoprecipitation”.
The LLM maps your phrase to actual database vocabulary (ChIP, ChIP-seq,
ChIP-qPCR) and ranks the researchers most strongly associated with each technique
in your matched papers.
How to read the result:
- Researcher card: name, paper count, year range, top techniques as green chips, broader L1 taxonomy chips below.
- Researcher profile: word clouds for Q topics / M techniques / F GO-tags; interactive treemaps; the full paper list (each clickable to the RPG graph); a Topic trajectory stacked bar chart per year; and Related researchers ranked by technique overlap.
- Browse All multi-select: type a technique name to add it as a chip; the list intersects (AND semantics) to find researchers who use all selected techniques.
Under the hood
- RPG extraction: each paper's text was processed by an LLM that emits the Q/M/F nodes and Q→M→F edges. The result is a per-paper directed acyclic graph.
- Taxonomy: a second LLM pass clusters nodes into L1/L2 categories per type. The taxonomy browser shows the tree; click any L1 to see its L2 children, papers, and researchers.
- Technique tags: a third LLM pass tags M-nodes with named techniques (PCR, ChIP-seq, …) keyed off the merged-CSV original text via MD5 hash. (Output has been audited; ~5% of tagger hallucinations removed.)
- Similarity (Related Papers): weighted Jaccard over the six taxonomy sets (Q·L1, Q·L2, M·L1, M·L2, F·L1, F·L2) per paper. Weights adjustable in the UI.
- Stack: FastAPI + Cytoscape.js + Anthropic Claude (Haiku for retrieval, Sonnet for the Digital Professor prospectus). Flat-file, no database.