Batch Polymer Embedding Analysis and Similarity Comparison

Scenario

A researcher needs to analyze 10 common polymer structures, generate TransPolymer embeddings for each, and identify which polymers are structurally most similar. The embeddings are then saved for use in a property prediction model.

Interactive Dialog

USER:

I have a list of 10 polymer structures I need to analyze. Can you generate embeddings for all of them and help me find which ones are most similar? Polyethylene: *CC* Polypropylene: *CC(C)* Polybutylene: *CC(CC)* Polystyrene: *CC(C1=CC=CC=C1)* PVC: *CC(Cl)* PVDF: *CC(F)(F)* PMMA: *CC(C)(C(=O)OC)* Nylon-6: *N(C(=O)CCCCC)C* PET: *OC(=O)C1=CC=C(C=C1)C(=O)OCC* Polycarbonate: *OC(C)(C)C1=CC=C(C=C1)C(C)(C)O*

ASSISTANT:

[Calls: brain_discovery]

ASSISTANT:

[Calls: brain_mount_service]

ASSISTANT:

[Calls: brain_predict]

Embedding Generation Results

All 10 polymers received 768-dimensional embeddings from TransPolymer.

Top 10 Most Similar Polymer Pairs

{ "pairs": [ { "polymer1": "PET", "polymer2": "Polycarbonate", "similarity": 0.9848 }, { "polymer1": "PMMA", "polymer2": "Nylon-6", "similarity": 0.9814 }, { "polymer1": "Polypropylene", "polymer2": "Polybutylene", "similarity": 0.9778 }, { "polymer1": "Polybutylene", "polymer2": "PVC", "similarity": 0.9764 }, { "polymer1": "Polypropylene", "polymer2": "PVC", "similarity": 0.9748 }, { "polymer1": "Nylon-6", "polymer2": "PET", "similarity": 0.9736 }, { "polymer1": "PMMA", "polymer2": "PET", "similarity": 0.9691 }, { "polymer1": "PMMA", "polymer2": "Polycarbonate", "similarity": 0.9686 }, { "polymer1": "Polystyrene", "polymer2": "PMMA", "similarity": 0.9652 }, { "polymer1": "Nylon-6", "polymer2": "Polycarbonate", "similarity": 0.9646 } ] }

Top 5 Least Similar Pairs

{ "pairs": [ { "polymer1": "Polyethylene", "polymer2": "Polycarbonate", "similarity": 0.7962 }, { "polymer1": "Polyethylene", "polymer2": "PET", "similarity": 0.8163 }, { "polymer1": "Polyethylene", "polymer2": "PMMA", "similarity": 0.8315 }, { "polymer1": "Polyethylene", "polymer2": "PVDF", "similarity": 0.842 }, { "polymer1": "Polyethylene", "polymer2": "Polystyrene", "similarity": 0.8445 } ] }

Key Insights

1. **PET ↔ Polycarbonate (98.5% similar)**: Both are aromatic polyesters with similar backbone structures containing benzene rings and ester linkages. 2. **Polyolefin cluster**: Polypropylene, Polybutylene, and PVC form a tight cluster (97-98% similar) - they share simple vinyl backbones. 3. **Engineering plastics cluster**: PMMA, Nylon-6, PET, and Polycarbonate are highly similar (96-98%) - all contain carbonyl functional groups. 4. **Polyethylene is the outlier**: Being the simplest structure, it's least similar to complex polymers like Polycarbonate (79.6%) and PET (81.6%).

USER:

Can you save these embeddings for later use in my property prediction model?

ASSISTANT:

Files Created

| File | Format | Use Case | |------|--------|----------| | `polymer_embeddings.json` | JSON | Human-readable, includes metadata | | `polymer_embeddings.npy` | NumPy (10×768) | Direct ML model input | | `polymer_names.json` | JSON | Name/SMILES mapping |

Usage Example

```python import numpy as np import json # Load embeddings as feature matrix X = np.load('polymer_embeddings.npy') # Shape: (10, 768) # Load names for reference with open('polymer_names.json') as f: mapping = json.load(f) names = mapping['names'] smiles = mapping['smiles'] # Use in your model # model.fit(X, y_properties) ```

Learning Outcomes

Generate TransPolymer embeddings for multiple polymers in batch

Successfully processed 10 polymers with 768D embeddings each

Calculate pairwise cosine similarity between polymer embeddings

Identified most and least similar polymer pairs

Interpret similarity results in chemical context

Explained structural reasons for similarity patterns

Export embeddings for downstream ML tasks

Saved in JSON and NumPy formats for property prediction

Key Concepts

TransPolymer Embeddings

768-dimensional vector representations that encode polymer structure and properties learned from large datasets

Cosine Similarity

Metric ranging from -1 to 1 measuring the angular similarity between two vectors; 1 = identical direction

Polymer SMILES

Simplified notation for polymer repeat units using * to denote connection points

Polyolefins

Simple hydrocarbon polymers (PE, PP, PB) with vinyl backbones

Engineering Plastics

High-performance polymers (PMMA, PC, PET, Nylon) typically containing carbonyl groups

Batch Processing

Generating embeddings for multiple structures in a single workflow

Tools Working

OK brain_discovery Found TransPolymer.exe available for mounting
OK brain_mount_service Mounted TransPolymer in ~45 seconds
OK brain_predict Generated 10 embeddings successfully