ETHER

Emerging Topic Horizon Engine for Research

ETHER is an open-source literature intelligence platform that turns any research topic into a live literature radar. Users define a topic, choose literature sources, configure priority signals, and generate ranked dashboards to track emerging research across biomedical science and beyond.

Why ETHER?

Researchers are overwhelmed by new papers. Standard alerts return long lists of results, but they rarely explain which papers matter most or why.

ETHER helps users:

• track new papers across topics
• prioritize papers using transparent scoring
• create reusable radar configurations
• monitor emerging research trends
• export literature tables for grants, reviews, and journal clubs
• share or clone dashboards

How It Works

Topic
  |
  v
Search sources
  |
  v
Extract metadata and abstracts
  |
  v
Apply signal profile
  |
  v
Calculate Research Relevance Score
  |
  v
Generate dashboard and exports

Core Concepts

Radar

A radar is a saved literature monitor for a research topic.

Examples:

• Cancer Evolution Radar
• AI Drug Discovery Radar
• Urban Microbiome Radar
• Single-Cell Immunology Radar

Signal Profile

A signal profile tells ETHER what to prioritize.

Example signal categories:

• methods
• clinical relevance
• datasets
• mechanisms
• biomarkers
• translational potential
• surveillance relevance
• therapeutic relevance
• open-source tools
• review articles

Research Relevance Score

The Research Relevance Score ranks papers using a transparent weighted model.

Research Relevance Score =
topic relevance
+ recency
+ priority term matches
+ method signals
+ source metadata
+ optional user feedback

Each paper card explains its score with the matched topic, priority terms, method signals, recency, source, and score components.

Product Workflow

ETHER is designed as a radar builder:

Create Radar -> Enter Topic -> Choose Template -> Adjust Signals -> Generate Dashboard -> Export

Users can:

1. Create a radar from a topic.
2. Select sources.
3. Define priority terms.
4. Set scoring weights.
5. Generate a live dashboard.
6. Save, clone, export, or publish it.

The fastest publishable version is CLI + configs + static dashboard + examples

• validation. User accounts, paid plans, team collaboration, and LLM summaries can come later.

Live Dashboard

https://aglucaci.github.io/ETHER/

Site Map

The static site now supports the MVP product and paper narrative:

• / - home page and live generated dashboard
• /create/ - guided radar builder
• /radar/cancer-evolution/ - example radar dashboard page
• /templates/ - reusable starter radars
• /explore/ - public radar gallery with clone actions
• /examples/ - finished dashboards and links to configs
• /methodology/ - Research Relevance Score explanation
• /docs/ - usage and reproducibility documentation
• /export/ - Research Pack export formats
• /about/ - project purpose, audience, and open-source framing

Quick Start

Current prototype:

pip install requests
python scripts/daily_pubmed_watch_v2.py --days 1 --max 12

Output:

docs/index.html
docs/latest.json
docs/latest.md

Run tests:

python -m unittest tests.test_scoring

Planned package workflow:

git clone https://github.com/your-org/ether
cd ether
pip install -e .
ether init ai_protein_design
ether build radars/ai_protein_design.yml
ether serve results/ai_protein_design/
ether export results/ai_protein_design/ --format csv
ether export results/ai_protein_design/ --format bibtex

Create Your Own Radar

Step 1: Choose a Topic

Example:

AI for protein design

Step 2: Start from a Template

Starter radar configurations live in examples/.

Recommended first templates:

• Cancer Evolution Radar
• Viral Evolution Radar
• Metagenomics Radar
• Urban Microbiome Radar
• AI Drug Discovery Radar
• Protein Design Radar
• Wastewater Surveillance Radar
• Single-Cell Genomics Radar

Step 3: Edit Your Query Set

radar_name: AI Protein Design Radar

sources:
  - pubmed
  - biorxiv
  - medrxiv

queries:
  - "protein design"
  - "protein engineering"
  - "generative model"
  - "diffusion model"
  - "protein language model"

Step 4: Define Your Signal Profile

signal_profile:
  methods:
    description: "Computational or experimental methods"
    terms:
      - diffusion model
      - protein language model
      - inverse folding
      - generative model
    weight: 2.0

  validation:
    description: "Experimental validation or benchmark evidence"
    terms:
      - wet-lab validation
      - binding assay
      - functional assay
      - benchmark
    weight: 2.5

  applications:
    description: "Applied protein engineering outcomes"
    terms:
      - enzyme design
      - antibody design
      - therapeutic protein
      - stability
    weight: 1.5

Step 5: Build, View, and Publish

ether build radars/ai_protein_design.yml
ether serve results/ai_protein_design/
ether publish results/ai_protein_design/

Example Radar Configuration

radar_name: Cancer Evolution Radar

description: >
  Tracks papers related to cancer evolution, clonal dynamics,
  therapy resistance, tumor phylogenetics, and single-cell methods.

sources:
  - pubmed
  - biorxiv
  - medrxiv

queries:
  - "cancer evolution"
  - "clonal evolution"
  - "tumor phylogeny"
  - "therapy resistance"
  - "subclonal architecture"

filters:
  date_range: "last_90_days"
  language: "english"

signal_profile:
  biological_process:
    terms:
      - clonal selection
      - subclonal expansion
      - tumor evolution
      - lineage tracing
    weight: 2.0

  methods:
    terms:
      - single-cell sequencing
      - phylogenetic reconstruction
      - mutational signatures
      - spatial transcriptomics
    weight: 1.5

  clinical_relevance:
    terms:
      - relapse
      - minimal residual disease
      - treatment resistance
      - metastasis
    weight: 2.5

scoring:
  topic_relevance_weight: 0.35
  recency_weight: 0.20
  signal_weight: 0.35
  source_weight: 0.10

outputs:
  - dashboard
  - csv
  - bibtex
  - markdown

Dashboard Features

Each dashboard includes:

• ranked papers
• Research Relevance Score
• score explanations
• matched priority terms
• source filters
• date filters
• emerging trend summaries
• saved papers
• export tools

Example paper card:

Title: Clonal dynamics during therapy resistance in lung cancer

Research Relevance Score: 91

Why ranked highly:
- matched topic: therapy resistance
- matched priority signal: clonal selection
- matched method: single-cell sequencing
- matched clinical term: relapse
- recent publication: 14 days old

Clone an Existing Radar

Users can clone a radar and modify it.

ether clone examples/cancer_evolution.yml my_leukemia_radar.yml

Then edit:

radar_name: Leukemia Evolution Radar

queries:
  - "acute myeloid leukemia"
  - "clonal hematopoiesis"
  - "therapy resistance"
  - "relapse"

Build:

ether build my_leukemia_radar.yml

User Levels

Level 1: Web User

For non-technical users:

Enter topic -> choose template -> adjust sliders -> generate dashboard

Level 2: Config User

For scientists who want reproducibility:

ether build my_topic.yml

This is ideal for labs, papers, and GitHub workflows.

Level 3: Developer User

For institutions and advanced users:

• add new source adapters
• add new scoring modules
• deploy a private instance
• connect internal databases
• add institutional templates
• run scheduled GitHub Actions

Repository Structure

.
|-- examples/
|   |-- cancer_evolution.yml
|   |-- viral_evolution.yml
|   |-- metagenomics.yml
|   |-- urban_microbiome.yml
|   |-- ai_drug_discovery.yml
|   |-- protein_design.yml
|   |-- wastewater_surveillance.yml
|   `-- single_cell_genomics.yml
|-- scripts/
|   `-- daily_pubmed_watch_v2.py
|-- logo/
|   `-- ETHER_logo.png
|-- docs/
|   `-- index.html
|-- tests/
|   `-- test_scoring.py
|-- manuscript/
|   `-- ether_manuscript.md
|-- MANUSCRIPT_PLAN.md
`-- .github/workflows/
    `-- ether-radar.yml

Automation

ETHER runs via GitHub Actions:

• daily radar update
• weekly deep scan
• automatic commit only when output changes
• concurrency-safe
• dependency-cached for speed

Workflow:

.github/workflows/ether-radar.yml

Validation Plan

For the paper, the key missing piece is validation.

Minimum evaluation:

• use cancer evolution, viral evolution, and metagenomics case studies
• retrieve 100-300 papers per case study
• manually label relevance
• compare ETHER ranking against baselines
• report Precision@10, Precision@25, nDCG@10, and Recall@50

Baselines:

• date-only ranking
• keyword-count ranking
• PubMed relevance ranking
• ETHER full model

Modest claim:

ETHER improves prioritization of expert-relevant papers compared with date-only and keyword-count baselines.

ETHER is not a replacement for systematic review. It is a surveillance and prioritization layer.

Paper Framing

Working title:

ETHER: Emerging Topic Horizon Engine for Research

Core novelty statement:

ETHER introduces a configurable signal-weighted framework for scientific literature surveillance, allowing users to define not only which topics to search, but which types of evidence, methods, applications, and emerging research signals should influence paper prioritization.

Roadmap

• PubMed search backend
• YAML radar config
• scoring model
• ranked CSV output
• static dashboard
• example radars
• BibTeX/RIS export
• GitHub Pages publishing
• web interface
• saved radars and user accounts
• full-text support
• LLM-assisted structured summaries
• user feedback learning
• team/shared radars
• citation graph integration

License

MIT License