Runs automated hypothesis generation and empirical testing on tabular datasets.

Install

mkdir -p .claude/skills/alterlab-hypogenic && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/16034" && unzip -o skill.zip -d .claude/skills/alterlab-hypogenic && rm skill.zip

Installs to .claude/skills/alterlab-hypogenic

Activation

This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.

Runs automated LLM-driven hypothesis generation and testing on tabular datasets with HypoGeniC, combining literature insights with data-driven testing. Use when systematically exploring hypotheses about patterns in empirical data (for example deception detection or content analysis). For manual hypothesis formulation use alterlab-hypothesis-gen; for open-ended creative ideation use alterlab-scientific-brainstorm. Part of the AlterLab Academic Skills suite.
460 chars✓ has a “when” triggerlonger than Claude Code's old 250-char listing cap (fine on current versions)
Intermediate

Key capabilities

  • →Generate scientific hypotheses from observational datasets.
  • →Test multiple competing hypotheses systematically.
  • →Combine literature insights with empirical patterns.
  • →Accelerate research discovery through automated hypothesis ideation.
  • →Extract insights from research papers via PDF processing.
  • →Prune near-duplicate hypotheses with a redundancy checker.

How it works

The skill uses LLMs to generate hypotheses from data, refines them based on performance, and can integrate insights from literature. It supports different methods like HypoGeniC, HypoRefine, and Union methods.

Inputs & outputs

You give it
Tabular datasets and optionally research papers.
You get back
Testable hypotheses, inference results, and performance metrics.

When to use alterlab-hypogenic

  • →Data-driven hypothesis generation
  • →Empirical data analysis
  • →Testing patterns in tabular data

About this skill

Hypogenic

Overview

Hypogenic provides automated hypothesis generation and testing using large language models to accelerate scientific discovery. The framework supports three approaches: HypoGeniC (data-driven hypothesis generation), HypoRefine (synergistic literature and data integration), and Union methods (mechanistic combination of literature and data-driven hypotheses).

Keyless by default: Claude runs the HypoGeniC loop

The large language model that HypoGeniC needs is Claude — so the whole loop runs in this conversation with no package and no API key. The deterministic scoring is a bundled stdlib script; Claude does the generate/refine reasoning. This is the default path; the hypogenic package (below) is an optional scaled-up alternative.

The loop (the HypoGeniC method, data-driven):

  1. Generate. Claude reads the labelled training split and proposes candidate hypotheses, each written as a checkable keyword rule — e.g. "texts containing any of {furthermore, moreover, additionally, consequently} are AI-generated". Record them in a small JSON bank (scripts/score_hypotheses.py --emit-template prints the shape).
  2. Score. Run the bundled, LLM-free scorer on a held-out split:
    python scripts/score_hypotheses.py \
      --data your_labelled_data.jsonl --hypotheses bank.json --split test
    
    It reports each hypothesis's held-out accuracy, the best single hypothesis, the first-match bank accuracy, and the examples the bank misclassifies. Pure stdlib, fully reproducible — no model call.
  3. Refine. Claude reads the scores and the misclassified examples, drops or edits weak hypotheses, and adds new ones aimed at the errors; re-score. Repeat until held-out accuracy plateaus (HypoGeniC's iterative refinement, done keyless).
  4. Report. Hand back the final hypothesis bank and its held-out accuracy, with the evidence (per-hypothesis and bank numbers from the scorer).

A worked example ships as a fixture: evals/fixtures/ai_vs_human_texts.jsonl (24 labelled human-vs-AI texts, train/test split). On its test split the connector hypothesis scores 0.875 and a two-hypothesis bank reaches 1.000 — verified with the scorer, so the loop is checkable end to end without any credential.

For your own data, give the scorer a .jsonl/.csv with text, label, and split columns (or pass --text-field/--label-field/--split-field). When features are not keyword-expressible, Claude can still apply each natural-language hypothesis to each held-out example to emit a predicted label, then compute accuracy the same deterministic way.

Scaled-up option: the hypogenic package

The upstream hypogenic package automates the same loop against an LLM provider and adds HypoRefine (literature) and Union methods. It needs its own environment and an LLM provider API key, and is worth it mainly for large banks, HypoRefine's PDF/literature integration, or heavy batch inference. Everything below documents that optional package.

Quick Start (optional package)

Get started with the hypogenic package in minutes:

# Install into its own environment: hypogenic 0.3.5 pins numpy~=1.26, pandas~=2.1,
# torch~=2.4, transformers~=4.45, openai~=1.40 and anthropic~=0.32 (Python 3.10–3.12)
uv venv .venv-hypogenic --python 3.10 && source .venv-hypogenic/bin/activate
uv pip install hypogenic

# Clone example datasets
git clone https://github.com/ChicagoHAI/HypoGeniC-datasets.git ./data

# Run data-driven hypothesis generation (model_type: gpt | claude | vllm | huggingface)
hypogenic_generation --task_config_path ./data/your_task/config.yaml \
    --model_type gpt --model_name <provider-model-id> \
    --max_num_hypotheses 20 --output_folder ./outputs/your_task

# Run inference with a saved hypothesis bank
hypogenic_inference --task_config_path ./data/your_task/config.yaml \
    --hypothesis_file ./outputs/your_task/<hypotheses_*.json> \
    --model_type gpt --model_name <provider-model-id>

Flags verified against the hypogenic 0.3.5 console scripts (hypogenic_cmd). The CLI covers data-driven HypoGeniC generation and inference; HypoRefine and Union runs are driven by the example scripts (below). Run --help for the full option list.

Or use the example scripts (the library ships runnable scripts under examples/; there is no one-line fluent task.generate_hypotheses(...) API — see "Python API Usage" below for the real classes):

python ./examples/generation.py   --help   # HypoGeniC data-driven generation
python ./examples/inference.py     --help   # single-hypothesis inference

When to Use This Skill

Use this skill when working on:

  • Generating scientific hypotheses from observational datasets
  • Testing multiple competing hypotheses systematically
  • Combining literature insights with empirical patterns
  • Accelerating research discovery through automated hypothesis ideation
  • Domains requiring hypothesis-driven analysis: deception detection, AI-generated content identification, mental health indicators, predictive modeling, or other empirical research

Does NOT Trigger

ScenarioUse Instead
Hand-formulating falsifiable hypotheses and predictions from observations (no LLM pipeline)alterlab-hypothesis-gen
Open-ended ideation or research-gap brainstorming with no dataset yetalterlab-scientific-brainstorm
Topic models, dictionaries, or supervised classifiers over text as social-science measurementalterlab-text-as-data
Testing one pre-specified hypothesis with a conventional statistical testalterlab-statistical-analysis

Hand-off rule. When the request matches a row above, open your answer by naming that skill by its exact id (for example alterlab-hypothesis-gen) and saying in one line why it fits better, then stop applying this skill's procedure. Name it even when you also need to ask for missing input. If only part of the request is out of scope, name the skill for that part and do only the in-scope part. A brief general pointer is fine; the named skill does the work.

Key Features

Automated Hypothesis Generation

  • Generate 10-20+ testable hypotheses from data in minutes
  • Iterative refinement based on validation performance
  • Support for both API-based (OpenAI, Anthropic) and local LLMs

Literature Integration

  • Extract insights from research papers via PDF processing
  • Combine theoretical foundations with empirical patterns
  • Systematic literature-to-hypothesis pipeline with GROBID

Performance Optimization

  • Redis caching reduces API costs for repeated experiments
  • Parallel processing for large-scale hypothesis testing
  • Adaptive refinement focuses on challenging examples

Flexible Configuration

  • Template-based prompt engineering with variable injection
  • Custom label extraction for domain-specific tasks
  • Modular architecture for easy extension

Reported Results (from arXiv:2410.17309, Literature Meets Data)

  • +8.97% over few-shot, +15.75% over literature-only, +3.37% over data-driven-only baselines
  • Human accuracy improved +7.44% (deception detection) and +14.19% (AI-generated content detection)
  • A redundancy checker prunes near-duplicate hypotheses to keep the final bank diverse

Core Capabilities

1. HypoGeniC: Data-Driven Hypothesis Generation

Generate hypotheses solely from observational data through iterative refinement.

Process:

  1. Initialize with a small data subset to generate candidate hypotheses
  2. Iteratively refine hypotheses based on performance
  3. Replace poorly-performing hypotheses with new ones from challenging examples

Best for: Exploratory research without existing literature, pattern discovery in novel datasets

2. HypoRefine: Literature and Data Integration

Synergistically combine existing literature with empirical data through an agentic framework.

Process:

  1. Extract insights from relevant research papers (typically 10 papers)
  2. Generate theory-grounded hypotheses from literature
  3. Generate data-driven hypotheses from observational patterns
  4. Refine both hypothesis banks through iterative improvement

Best for: Research with established theoretical foundations, validating or extending existing theories

3. Union Methods

Mechanistically combine literature-only hypotheses with framework outputs.

Variants:

  • Literature ∪ HypoGeniC: Combines literature hypotheses with data-driven generation
  • Literature ∪ HypoRefine: Combines literature hypotheses with integrated approach

Best for: Comprehensive hypothesis coverage, eliminating redundancy while maintaining diverse perspectives

Installation

Install via pip:

uv pip install hypogenic

Because of the pinned dependency versions above, keep hypogenic out of your main analysis environment.

Optional dependencies:

  • Redis server (port 6832): Enables caching of LLM responses to significantly reduce API costs during iterative hypothesis generation
  • s2orc-doc2json: Required for processing literature PDFs in HypoRefine workflows
  • GROBID: Required for PDF preprocessing (see Literature Processing section)

Clone example datasets:

# For HypoGeniC examples
git clone https://github.com/ChicagoHAI/HypoGeniC-datasets.git ./data

# For HypoRefine/Union examples
git clone https://github.com/ChicagoHAI/Hypothesis-agent-datasets.git ./data

Dataset Format

Datasets must follow HuggingFace datasets format with specific naming conventions:

Required files:

  • <TASK>_train.json: Training data
  • <TASK>_val.json: Validation data
  • <TASK>_test.json: Test data

Required keys in JSON:

  • text_features_1 through text_features_n: Lists of strings containing feature values
  • label: List of strings containing ground truth labels

Example (headline click prediction):

{
  "headline_1": [
    "What Up, Comet? You Just Got *PROBED*",


---

*Content truncated.*

When not to use it

  • →For manual hypothesis formulation, use alterlab-hypothesis-gen.
  • →For open-ended creative ideation, use alterlab-scientific-brainstorm.

Prerequisites

hypogenic Python packageLLM provider API key (e.g., OPENAI_API_KEY)

Limitations

  • →It requires an LLM provider API key for hypothesis generation.
  • →Optional dependencies like Redis server, s2orc-doc2json, and GROBID are needed for full functionality.
  • →It is not designed for open-ended creative ideation.

How it compares

This skill automates the entire hypothesis generation and testing process, combining data-driven and literature-based approaches, which is more systematic and efficient than manual hypothesis formulation.

Compared to similar skills

alterlab-hypogenic side by side with the closest alternatives in the catalog.

SkillInstallsUpdatedSafetyDifficulty
alterlab-hypogenic (this skill)04moReviewIntermediate
model-compare711moReviewAdvanced
cellxgene-census811moReviewAdvanced
tcga-bulk-data-preprocessing-with-omicverse18moNo flagsAdvanced

Try saying

Example prompts that trigger this skill in your AI assistant.

You might also like

model-compare

rawwerks

Compare 3D CAD models using boolean operations (IoU, Dice, precision/recall). Use when evaluating generated models against gold references, diffing CAD revisions, or computing similarity metrics for ML training. Triggers on: model diff, compare models, IoU, intersection over union, model similarity, CAD comparison, STEP diff, 3D evaluation, gold reference, generated model, precision recall 3D.

783

cellxgene-census

davila7

Query CZ CELLxGENE Census (61M+ cells). Filter by cell type/tissue/disease, retrieve expression data, integrate with scanpy/PyTorch, for population-scale single-cell analysis.

88

tcga-bulk-data-preprocessing-with-omicverse

Starlitnightly

Guide Claude through ingesting TCGA sample sheets, expression archives, and clinical carts into omicverse, initialising survival metadata, and exporting annotated AnnData files.

14

torchdrug

davila7

Graph-based drug discovery toolkit. Molecular property prediction (ADMET), protein modeling, knowledge graph reasoning, molecular generation, retrosynthesis, GNNs (GIN, GAT, SchNet), 40+ datasets, for PyTorch-based ML on molecules, proteins, and biomedical graphs.

326

string-database

davila7

Query STRING API for protein-protein interactions (59M proteins, 20B interactions). Network analysis, GO/KEGG enrichment, interaction discovery, 5000+ species, for systems biology.

217

transformer-lens-interpretability

davila7

Provides guidance for mechanistic interpretability research using TransformerLens to inspect and manipulate transformer internals via HookPoints and activation caching. Use when reverse-engineering model algorithms, studying attention patterns, or performing activation patching experiments.

215

Search skills

Search the agent skills registry