TADF Screening Skill

Standardized, stage-gated workflow for high-throughput luminescent materials screening (TADF, Fluorescence, Phosphorescence).

This skill is designed for autonomous agents and human-in-the-loop execution. It enforces strict boundary initialization, topology-aware assembly, stability filtering, wavelength filtering, and final TDDFT validation.

OpenClaw-Powered Agentic Workflow

Comparison of traditional manual TADF screening and the OpenClaw-powered agentic workflow used in this project.

---

What this skill does

• Initializes workflow boundary conditions before any generation starts.
• Builds donor/acceptor candidate molecules with strict topology rules.
• Runs structural stability filtering with xTB.
• Runs optional TDDFT-xTB wavelength-window filtering.
• Prepares elite candidates for full TDDFT validation.
• Enforces stage gates and inquiry-stage clarification rules.

---

Folder structure (standard)

skills/tadf-screening/
├── SKILL.md
├── README.md
├── BATCH_STATUS.md
├── data/
│   └── da_library_30.csv
├── examples/
│   └── molecules.csv
├── references/
│   ├── workflow-spec.md
│   └── io-contract.md
└── scripts/
    ├── screening_workflow_initializer.py
    ├── build_da_topology_library.py
    ├── run_xtb_batch_manifest.py
    ├── run_tddft_xtb_filter.py
    ├── smiles_assembler.py
    └── molzip_assembler.py

---

Module-by-module description

1) scripts/screening_workflow_initializer.py

Foundational initializer. Must run first.

Responsibilities:

• Detect hardware (cpu, ram, gpu) and Slurm/HPC hints.
• Assign compute_tier: local_basic | local_gpu | remote_cluster.
• Capture photophysical target constraints:
  • emission_range_nm
  • emission_type
  • spectrum_width_requirement
  • empirical_stokes_shift_ev
• Discover TDDFT engines (gaussian, orca, qchem, pyscf) and apply smart defaults.
• Enforce active inquiry mode for missing critical fields.
• Output JSON configuration for downstream stages.

Critical protocol enforced in initializer:

• Do not estimate emission from S0-only vertical excitation.
• Use: S0 optimization -> excited-state optimization (S1/T1) -> vertical emission.

---

2) scripts/build_da_topology_library.py

Topology-aware D/A assembly script.

Supported topologies:

• D-A
• D-A-D
• A-D-A
• D-pi-A
• D_n-A

Key constraints:

• Uses RDKit assembly logic (molzip-mapped fragments; compatible with reaction-style workflows).
• For D-A-D, acceptor must have >=2 leaving groups (Cl/Br/I) or candidate is skipped.
• Accepts initial sample count (--initial-sample-count, default 10000).

Output:

• CSV with assembled product_smiles candidates.

---

3) scripts/run_xtb_batch_manifest.py

Batch xTB pre-screening runner.

Responsibilities:

• Read manifest (idx,name,xyz_path).
• Execute xTB per molecule.
• Track checkpoints and per-sample status.

Output contracts:

• xtb_state.json (resume state)
• xtb_progress.csv (detailed per-sample result)

Typical usage in this workflow:

• Structural stability and coarse electronic pre-filtering.

---

4) scripts/run_tddft_xtb_filter.py

Semi-empirical wavelength-window filter after xTB.

Responsibilities:

• Run TDDFT-xTB-like stage (depending on environment tools).
• Parse excitation proxy.
• Apply empirical Stokes shift.
• Keep molecules inside target emission window.

Output:

• tddft_xtb_results.csv
• tddft_xtb_blue_window.csv

TADF optional gate:

• Apply practical ΔE_ST threshold rule when available.

---

5) scripts/smiles_assembler.py and scripts/molzip_assembler.py

SMILES-level assembly and validation tools.

Responsibilities:

• Fragment validation and auditable rejection reasons.
• Connectivity and sanitization checks.
• Mapped-fragment assembly via molzip route when needed.

---

Quick start by scenarios

Scenario A: Blue TADF (default practical path)

1. Run initializer with emission_type=TADF, emission_range_nm=450-490, narrow width preference.
2. Select topology (D-A-D or mixed) and sample count (default 10000).
3. Build topology library (build_da_topology_library.py).
4. Generate .xyz structures with your project generator.
5. Run run_xtb_batch_manifest.py.
6. Run run_tddft_xtb_filter.py with your empirical Stokes shift.
7. Send final shortlist to full TDDFT (S0 -> S1/T1 -> emission).

Scenario B: Fluorescence emitters (singlet-focused)

1. Set emission_type=Fluorescence.
2. Keep tighter oscillator/emission filtering in Stage 2.
3. Final validation must include S1 excited-state optimization before emission.

Scenario C: Phosphorescence emitters (triplet-focused)

1. Set emission_type=Phosphorescence.
2. Use topology and acceptor choices suitable for stronger SOC pathways.
3. Final validation must include T1 optimization before emission.

Scenario D: Proprietary fragment libraries

1. Keep built-in DeepChem/PubChem enabled.
2. Add custom_db_paths (.csv / .smi) in initializer.
3. Re-run topology assembly on merged library.

---

Stage-by-stage screening protocol

Stage 0 — Initialization (mandatory)

Run initializer and resolve missing critical fields.

Stage 1 — Topology assembly + RDKit/MMFF94 + xTB stability

• Build candidates by topology.
• Generate 3D structures.
• Run xTB structure optimization/stability checks.
• Discard non-converged or structurally unstable molecules.

Stage 2 — sTD-DFT/xTB wavelength filtering

• Compute absorption proxies.
• Apply empirical_stokes_shift_ev.
• Keep candidates within target emission range.
• For TADF, apply ΔE_ST gate if available.

Stage 3 — Full TDDFT validation

• Elite candidates only.
• Protocol: S0 opt -> S1/T1 opt -> vertical emission.
• Use selected engine and tddft_level from initializer.

Stage 4 — MOMAP Photophysics (NEW ✨)

• Engine: MOMAP 2024A (TVCF method, B3LYP/6-31G*).
• Requires: Gaussian .log + .fchk from Stage 3 (S0/S1/T1).
• Computes:
  • EVC (electron-vibration coupling): Duschinsky matrix + Huang-Rhys factors.
  • spec_tvcf: Franck-Condon emission spectrum with temperature broadening.
  • ISC rate: S1→T1 intersystem crossing rate (key for TADF RISC).
• Filter: peak emission in blue window (450–490 nm) + ΔE_ST < 0.3 eV.
• Output: stage4_report.md (ranked), stage4_results.json, per-molecule spectrum PNG.

Usage:

# Single molecule
python scripts/stage4_momap.py --mol-id mol_07566 \
    --s0 logs/s0.log --s1 logs/s1.log --t1 logs/t1.log

# Batch from CSV (cols: mol_id,s0_log,s1_log,t1_log)
python scripts/stage4_momap.py candidates.csv --output stage4_output/

Scoring: blue window proximity × ΔE_ST × f_emi → ranked shortlist.

---

Stage-gate policy

• If Stage 1 yields ok=0: stop and diagnose before Stage 2.
• If required tools for Stage 2 are missing: emit explicit tool_missing and do not fake results.
• If remote execution is used, batch metadata must record host/workdir/pid/timestamps.

---

References

• references/workflow-spec.md
• references/io-contract.md