Scientific Hypothesis Generation

Systematically develop testable explanations and experimental designs.

When to Use

• Developing hypotheses from observations or preliminary data

• Filling out .research/project_telos.md aims section

• During the PLANNING phase

• Exploring competing explanations for phenomena

• Designing experiments to test research questions

• Generating predictions for research proposals

Workflow

1. UNDERSTAND → Clarify the phenomenon/question
2. RESEARCH → Survey existing literature
3. SYNTHESIZE → Integrate evidence and identify gaps
4. GENERATE → Develop 3-5 competing hypotheses
5. EVALUATE → Assess hypothesis quality
6. DESIGN → Plan experimental tests
7. PREDICT → Formulate testable predictions

Step 1: Understand the Phenomenon

Clarifying Questions

Ask these to define the research question:

What is the core observation or pattern?

• What specifically needs to be explained?

What is the scope?

• What are the boundaries of the phenomenon?

• What is included/excluded?

What is already known?

• Established facts vs. assumptions

• Previous attempts at explanation

What are the constraints?

• Methodological limitations

• Time/resource constraints

• Ethical considerations

Example

Phenomenon Definition

Observation: Treatment X reduces tumor growth in mice, but only in animals with intact immune systems.

Scope: Focus on solid tumors in murine models. Excludes metastasis and blood cancers.

Known: Treatment X has no direct cytotoxic effect on cancer cells in vitro.

Question: How does Treatment X reduce tumor growth in an immune-dependent manner?

Step 2: Literature Research

Before generating hypotheses, ground them in existing evidence.

Search Strategy

• Identify key concepts from the research question

• List synonyms and related terms

• Search systematically (use /deep_research if needed)

• Document findings for later reference

Integration with RA

Use /deep_research to gather literature:

/deep_research Treatment X mechanism of action cancer immunity

This creates documented summaries in .research/literature/

Step 3: Synthesize Existing Evidence

Evidence Summary Template

Literature Synthesis

What is Established

1. [Established fact 1 with citation]
2. [Established fact 2 with citation]

Current Theories

1. [Theory A]: [Brief description] (Supporting: [refs], Contradicting: [refs])
2. [Theory B]: [Brief description]

Knowledge Gaps

1. [Gap 1]: No studies have examined...
2. [Gap 2]: Conflicting results regarding...

Relevant Mechanisms from Related Systems

1. [Analogous system 1]: [What can be learned]
2. [Analogous system 2]: [Potential parallel]

Step 4: Generate Competing Hypotheses

Develop 3-5 distinct hypotheses that could explain the phenomenon.

Hypothesis Requirements

Each hypothesis must be:

• Mechanistic: Explains how and why, not just what

• Testable: Can be evaluated empirically

• Distinguishable: Different from other hypotheses in testable ways

• Evidence-based: Grounded in existing knowledge

Strategies for Generation

Strategy Description Example

Analogical Apply mechanisms from similar systems "Similar to how X works in Y system"

Mechanistic decomposition Break down into component processes "Step 1 leads to Step 2 which causes..."

Level shifting Consider different scales "At the molecular level..." vs "At the systems level..."

Contradiction exploration What if the opposite were true? "What if X inhibits rather than activates?"

Integration Combine known mechanisms in new ways "If A and B act together..."

Example Hypotheses

Competing Hypotheses

H1: T-cell Activation Hypothesis

Treatment X enhances T-cell activation through direct binding to checkpoint receptors, leading to increased tumor infiltration and cytotoxicity.

Mechanism: X → checkpoint binding → T-cell activation → tumor killing Key prediction: T-cell depletion would abolish the effect

H2: Dendritic Cell Priming Hypothesis

Treatment X stimulates dendritic cell maturation and antigen presentation, leading to enhanced adaptive immune response against tumor antigens.

Mechanism: X → DC maturation → antigen presentation → T-cell priming Key prediction: Effect would require intact antigen presentation

H3: Tumor Microenvironment Remodeling Hypothesis

Treatment X alters the immunosuppressive tumor microenvironment by depleting regulatory T cells or MDSCs, allowing existing immune responses to act.

Mechanism: X → Treg/MDSC depletion → reduced immunosuppression Key prediction: Effect correlates with reduction in immunosuppressive cells

H4: Innate Immune Activation Hypothesis

Treatment X activates innate immune cells (NK cells, macrophages) that directly kill tumor cells and enhance adaptive immunity.

Mechanism: X → innate activation → tumor killing + cytokine release Key prediction: Early innate response precedes adaptive response

Step 5: Evaluate Hypothesis Quality

Quality Criteria

Criterion Question Score (1-5)

Testability Can it be empirically tested?

Falsifiability What would disprove it?

Parsimony Is it the simplest explanation?

Explanatory power How much does it explain?

Scope What range of observations does it cover?

Consistency Does it fit established principles?

Novelty Does it offer new insights?

Evaluation Template

Hypothesis Evaluation

Strongest hypothesis: H1 (most testable, clear predictions) Alternative to test: H3 (could explain H1 results)

Step 6: Design Experimental Tests

For Each Hypothesis, Define:

• Key experiment: The critical test

• Controls: What comparisons are needed

• Methods: How would you measure outcomes

• Expected results: If hypothesis is correct

• Alternative outcomes: What other results could mean

Experimental Design Template

Experimental Design: Testing H1

Critical Experiment

Deplete CD8+ T cells using anti-CD8 antibodies before Treatment X administration.

Experimental Groups

1. Treatment X + isotype control antibody (n=10)
2. Treatment X + anti-CD8 antibody (n=10)
3. Vehicle + isotype control (n=10)
4. Vehicle + anti-CD8 antibody (n=10)

Primary Outcome

Tumor volume at day 14 post-treatment

Expected Results (if H1 correct)

• Group 1 shows reduced tumor growth
• Group 2 shows NO reduction (similar to Group 3)
• This would demonstrate T-cell dependence

Alternative Outcomes

• If Group 2 still shows reduction → Effect is T-cell independent → Support for H3 or H4
• If Group 4 shows increased growth → T cells contribute to baseline control → Consider immunocompetent models

Step 7: Formulate Testable Predictions

Prediction Requirements

Good predictions are:

• Specific: Clear, measurable outcomes

• Quantitative (when possible): Expected magnitude or direction

• Conditional: Specify under what conditions

• Distinguishing: Differentiate between hypotheses

Prediction Format

Predictions

If H1 is correct:

1. CD8+ T cell infiltration will increase >2-fold after Treatment X
2. T cell depletion will abolish >80% of tumor reduction effect
3. PD-1/PD-L1 blockade will enhance Treatment X efficacy synergistically

If H2 is correct:

1. Dendritic cell maturation markers (CD80, CD86) will increase
2. Antigen presentation blockade will eliminate the effect
3. The effect will require 7+ days (time for adaptive response)

Discriminating Predictions:

• H1 predicts rapid effect (days); H2 predicts delayed effect (weeks)
• H1 predicts T cell depletion is sufficient; H2 predicts DC depletion is also required

Integration with RA Workflow

Output to project_telos.md

The generated hypotheses should be added to .research/project_telos.md :

Aim 1: Determine the mechanism of Treatment X efficacy

• Hypothesis: Treatment X enhances T-cell activation through checkpoint receptor binding, leading to increased tumor infiltration and cytotoxicity. (H1 from hypothesis generation)
• Alternative: The effect may be mediated by tumor microenvironment remodeling (H3).
• Approach: T-cell depletion experiments followed by immune profiling
• Success Criteria: Identify the critical immune cell population required for Treatment X efficacy
• Status: Not started

Phase Gate Contribution

This skill helps complete the PLANNING phase requirements:

• Project aims are defined ← Hypothesis generation contributes here

• At least one literature search completed

• background.md has at least a rough draft

Hypothesis Generation Checklist

• Phenomenon clearly defined and bounded

• Literature searched and synthesized

• 3-5 competing hypotheses generated

• All hypotheses are mechanistic and testable

• Quality evaluation completed

• Experimental designs outlined

• Predictions formulated and distinguish between hypotheses

• Added to project_telos.md