Scientific Prediction Skill
Predict properties, trends, and outcomes across scientific disciplines.
When to Use
- "Predict the solubility of this compound"
- "What's the expected trend for..."
- "Estimate the effect size for this intervention"
- "Forecast the trajectory of..."
- "What properties would this material have?"
- Model-based estimation tasks
When NOT to Use
- Looking up known properties (use literature-search)
- Running actual computations (use code-execution)
- Verifying existing predictions (use scienceclaw-verification)
- Real-time data monitoring or alerts
Prediction Categories
1. Property Prediction
- Chemistry: Molecular properties (logP, solubility, toxicity, pKa, boiling point)
- Materials: Mechanical (strength, hardness), thermal, electrical properties
- Biology: Protein function, binding affinity, gene expression levels
- Physics: Material behavior under conditions (temperature, pressure)
2. Trend Analysis
- Time-series extrapolation with confidence intervals
- Growth/decay curve fitting (exponential, logistic, polynomial)
- Seasonal pattern identification
- Regime change detection
3. Outcome Prediction
- Clinical trial outcome estimation
- Experimental result prediction
- Treatment response probability
- Environmental impact forecasting
4. Model-Based Estimation
- QSAR/QSPR (quantitative structure-activity/property relationships)
- Pharmacokinetic modeling (ADME)
- Population dynamics modeling
- Economic indicator forecasting
Output Format
All predictions must include:
**Prediction**: [Value or range]
**Confidence Interval**: [Lower - Upper] at [confidence level]%
**Method**: [Approach used]
**Key Assumptions**: [List]
**Uncertainty Sources**: [List]
**Validation**: [How to verify this prediction]
**Caveats**: [Known limitations]
Guidelines
- Always quantify uncertainty — never provide point estimates without ranges
- State assumptions explicitly — hidden assumptions undermine predictions
- Distinguish extrapolation from interpolation — flag when predicting outside training data range
- Consider domain constraints — physical laws, biological limits, economic boundaries
- Recommend validation approaches — suggest experiments or data to verify predictions
- Use appropriate models — match model complexity to data availability
- Flag low-confidence predictions — be transparent about reliability
Discipline-Specific Methods
| Domain |
Common Methods |
| Chemistry |
QSAR, DFT calculations, molecular dynamics |
| Biology |
Sequence-based prediction, network analysis |
| Medicine |
Cox regression, Kaplan-Meier, NNT/NNH |
| Physics |
Theoretical models, scaling laws |
| Economics |
Econometric models, agent-based simulation |
| Climate |
GCM projections, statistical downscaling |
| Materials |
Phase diagrams, computational screening |
| Sociology |
Panel data models, social network evolution |
Computational Prediction Tools
When predictions require computation, integrate with these skills:
Molecular Property Prediction
- Use rdkit-chemistry for descriptor-based QSAR models
- Use pubchem-compound to retrieve experimental property values for training data
- Use scikit-learn-ml to build/evaluate prediction models
Materials Property Prediction
- Use materials-project to retrieve DFT-computed properties
- Use pymatgen-materials for structure-property analysis
- Use scipy-analysis for interpolation and regression
Biological Outcome Prediction
- Use biopython-bio for sequence-based feature extraction
- Use transformers-inference for protein language models (ESM, ProtTrans)
- Use scanpy-singlecell for cell-type and trajectory prediction
Geospatial/Climate Prediction
- Use geopandas-spatial for spatial feature engineering
- Use copernicus-climate for historical climate data as training input
- Use statsmodels-stats for time-series forecasting models
Zero-Hallucination Rule
ALL factual claims, citations, database results, and scientific data presented to the user MUST come from actual tool results (API calls, code execution, web search) in this conversation. NEVER fabricate or "fill in" details from training data. If a tool returns no results or partial data, report exactly what happened.
1---
2name: scienceclaw-prediction
3description: "Predict scientific properties, trends, and outcomes. Use when: user asks for property prediction, trend forecasting, or model-based estimation. NOT for: historical data lookup or real-time monitoring."
4metadata: { "openclaw": { "emoji": "📈" } }
5---
6
7# Scientific Prediction Skill
8
9Predict properties, trends, and outcomes across scientific disciplines.
10
11## When to Use
12
13- "Predict the solubility of this compound"
14- "What's the expected trend for..."
15- "Estimate the effect size for this intervention"
16- "Forecast the trajectory of..."
17- "What properties would this material have?"
18- Model-based estimation tasks
19
20## When NOT to Use
21
22- Looking up known properties (use literature-search)
23- Running actual computations (use code-execution)
24- Verifying existing predictions (use scienceclaw-verification)
25- Real-time data monitoring or alerts
26
27## Prediction Categories
28
29### 1. Property Prediction
30- **Chemistry**: Molecular properties (logP, solubility, toxicity, pKa, boiling point)
31- **Materials**: Mechanical (strength, hardness), thermal, electrical properties
32- **Biology**: Protein function, binding affinity, gene expression levels
33- **Physics**: Material behavior under conditions (temperature, pressure)
34
35### 2. Trend Analysis
36- Time-series extrapolation with confidence intervals
37- Growth/decay curve fitting (exponential, logistic, polynomial)
38- Seasonal pattern identification
39- Regime change detection
40
41### 3. Outcome Prediction
42- Clinical trial outcome estimation
43- Experimental result prediction
44- Treatment response probability
45- Environmental impact forecasting
46
47### 4. Model-Based Estimation
48- QSAR/QSPR (quantitative structure-activity/property relationships)
49- Pharmacokinetic modeling (ADME)
50- Population dynamics modeling
51- Economic indicator forecasting
52
53## Output Format
54
55All predictions must include:
56
57```
58**Prediction**: [Value or range]
59**Confidence Interval**: [Lower - Upper] at [confidence level]%
60**Method**: [Approach used]
61**Key Assumptions**: [List]
62**Uncertainty Sources**: [List]
63**Validation**: [How to verify this prediction]
64**Caveats**: [Known limitations]
65```
66
67## Guidelines
68
691. **Always quantify uncertainty** — never provide point estimates without ranges
702. **State assumptions explicitly** — hidden assumptions undermine predictions
713. **Distinguish extrapolation from interpolation** — flag when predicting outside training data range
724. **Consider domain constraints** — physical laws, biological limits, economic boundaries
735. **Recommend validation approaches** — suggest experiments or data to verify predictions
746. **Use appropriate models** — match model complexity to data availability
757. **Flag low-confidence predictions** — be transparent about reliability
76
77## Discipline-Specific Methods
78
79| Domain | Common Methods |
80|--------|---------------|
81| Chemistry | QSAR, DFT calculations, molecular dynamics |
82| Biology | Sequence-based prediction, network analysis |
83| Medicine | Cox regression, Kaplan-Meier, NNT/NNH |
84| Physics | Theoretical models, scaling laws |
85| Economics | Econometric models, agent-based simulation |
86| Climate | GCM projections, statistical downscaling |
87| Materials | Phase diagrams, computational screening |
88| Sociology | Panel data models, social network evolution |
89
90## Computational Prediction Tools
91
92When predictions require computation, integrate with these skills:
93
94### Molecular Property Prediction
95- Use **rdkit-chemistry** for descriptor-based QSAR models
96- Use **pubchem-compound** to retrieve experimental property values for training data
97- Use **scikit-learn-ml** to build/evaluate prediction models
98
99### Materials Property Prediction
100- Use **materials-project** to retrieve DFT-computed properties
101- Use **pymatgen-materials** for structure-property analysis
102- Use **scipy-analysis** for interpolation and regression
103
104### Biological Outcome Prediction
105- Use **biopython-bio** for sequence-based feature extraction
106- Use **transformers-inference** for protein language models (ESM, ProtTrans)
107- Use **scanpy-singlecell** for cell-type and trajectory prediction
108
109### Geospatial/Climate Prediction
110- Use **geopandas-spatial** for spatial feature engineering
111- Use **copernicus-climate** for historical climate data as training input
112- Use **statsmodels-stats** for time-series forecasting models
113
114## Zero-Hallucination Rule
115
116ALL factual claims, citations, database results, and scientific data presented to the user MUST come from actual tool results (API calls, code execution, web search) in this conversation. NEVER fabricate or "fill in" details from training data. If a tool returns no results or partial data, report exactly what happened.
beita6969 