Comparing Polymer Structures with AI

The BRAIN platform is running and ready to host AI models.

**TransPolymer** - Transformer-based model that analyzes polymer repeat units and creates a 768-number 'fingerprint' capturing chemical character. **PolyNC_API** - Neural computation model with similar capabilities. Think of these fingerprints like a detailed chemical profile - polymers with similar structures will have similar profiles.

✅ TransPolymer - running, healthy ✅ PolyNC_API - running, healthy Both models are ready to analyze your polymers.

Use SMILES notation for your repeat unit. Mark the chain connection points with asterisks (*). **Example:** Polyethylene repeat unit is `*CC*` - The two carbons (CC) are the repeat unit - The asterisks show where it connects to the next unit

| Polymer | Repeat Unit | SMILES | Category | |---------|-------------|--------|----------| | Polyethylene (PE) | –CH₂–CH₂– | `*CC*` | basic_polymer | | Polystyrene (PS) | –CH₂–CH(Ph)– | `*CC(c1ccccc1)*` | aromatic_polymer | | PMMA | –CH₂–C(CH₃)(COOCH₃)– | `*CC(C)(C(=O)OC)*` | ester_polymer | | PVC | –CH₂–CHCl– | `*CCCl*` | halogenated_polymer | | PEO/PEG | –CH₂–CH₂–O– | `*CCO*` | ether_polymer | | Nylon-6 | –NH–(CH₂)₅–CO– | `*NCCCCCC(=O)*` | amide_polymer |

TransPolymer generates **768-dimensional embeddings**. Typical computation time: **20-100ms**

**Result:** 768-dimensional fingerprint generated in 21ms **First 5 values:** [-0.120, -0.688, 0.028, 0.721, 0.936] **What the fingerprint captures:** - Simple saturated hydrocarbon backbone - No polar groups - No aromatic rings - Flexible chain structure

**Result:** 768-dimensional fingerprint generated in 21ms **First 5 values:** [-0.838, -0.676, 0.197, 0.066, 0.450] **What the fingerprint captures:** - Aromatic pendant group (benzene ring) - Bulky side chain restricts chain mobility - π-electron system from aromatic ring - Rigid compared to PE

| Dimension | PE | PS | Difference | |-----------|-----|-----|------------| | 1 | -0.120 | -0.838 | **0.718** | | 2 | -0.688 | -0.676 | 0.012 | | 3 | 0.028 | 0.197 | 0.169 | | 4 | 0.721 | 0.066 | **0.655** | | 5 | 0.936 | 0.450 | **0.486** | Dimensions 1, 4, and 5 show large differences - likely encoding the aromatic character of PS vs the simple alkane structure of PE.

**Result:** 768-dimensional fingerprint generated in 18ms **First 5 values:** [-0.430, -0.807, -0.500, -0.224, 0.575] **What the fingerprint captures:** - Ester group (–COOCH₃) provides polarity - Methyl groups on backbone and side chain - Good optical clarity (used in Plexiglass) - Moderate chain stiffness

| Dimension | PE | PS | PMMA | |-----------|------|------|------| | 1 | -0.120 | -0.838 | -0.430 | | 2 | -0.688 | -0.676 | -0.807 | | 3 | 0.028 | 0.197 | **-0.500** | | 4 | 0.721 | 0.066 | -0.224 | | 5 | 0.936 | 0.450 | 0.575 | **Observations:** - Dimension 3: PMMA is strongly negative (-0.500) vs PE/PS positive — likely encoding the **polar ester group** - Dimension 4: PE is high (0.721), PMMA is negative (-0.224) — may relate to **chain flexibility** - The fingerprints clearly distinguish all three polymer types

**Use case:** You have a polymer that works well but need alternatives. Compare fingerprints to find polymers with similar chemical character. Polymers with close fingerprints often share: - Similar Tg ranges - Compatible solubility - Related mechanical properties

**Use case:** Will two polymers blend well? Polymers with similar fingerprints tend to be more miscible. For example, our PE and PS fingerprints differ significantly in dimension 1 (-0.120 vs -0.838), which correlates with their known immiscibility.

**Use case:** You have 100 candidate monomers. Which are worth synthesizing? 1. Generate fingerprints for all candidates (takes ~20ms each) 2. Compare to your target polymer 3. Rank by fingerprint similarity 4. Focus synthesis on top candidates

**Use case:** Predict Tg from structure. Use the 768-dimensional fingerprints as input features: - Collect known Tg values for training polymers - Train regression model (fingerprint → Tg) - Predict Tg for new polymers from their fingerprints

**What we did:** 1. Checked BRAIN platform status and available models 2. Confirmed TransPolymer and PolyNC are running 3. Analyzed three polymers: - **PE** (simple, nonpolar, flexible) - **PS** (aromatic, rigid) - **PMMA** (polar ester, moderate stiffness) 4. Compared fingerprint values to understand structural differences **Key insight:** The 768-dimensional fingerprints successfully distinguish polymer types based on their chemical character — aromatic vs aliphatic, polar vs nonpolar, flexible vs rigid.

Try analyzing your own polymers: ``` brain_predict(service_name="TransPolymer.exe", input_data={"data": {"text": "YOUR_SMILES"}, "input_type": "text"}) ``` Remember: Use asterisks (*) to mark polymer chain connection points.