Official Resources
- Source Repository: https://github.com/grimme-lab/ml4nmr
- Documentation: Included in repository
- License: Open source
Overview
ml4nmr is a machine learning-based correction tool for NMR chemical shifts calculated with DFT. It enables correction of 1H and 13C NMR chemical shifts toward CCSD(T) quality (ΔcorrML) and prediction of spin-orbit relativistic contributions to NMR chemical shifts caused by heavy atoms (ΔSO^ML).
Scientific domain: ML-corrected NMR chemical shifts, spin-orbit corrections
Target user community: Researchers needing high-accuracy NMR chemical shift predictions including heavy-atom relativistic effects
Theoretical Methods
- Machine learning correction of DFT NMR shifts (ΔcorrML)
- CCSD(T) quality from DFT+ML
- Spin-orbit relativistic correction (ΔSO^ML)
- 1H and 13C chemical shift correction
- Heavy atom effects on chemical shifts
- Grimme lab DFT baseline (XTB)
Capabilities (CRITICAL)
- ML correction of DFT NMR chemical shifts
- ΔcorrML: DFT→CCSD(T) quality correction
- ΔSO^ML: Spin-orbit relativistic correction
- 1H and 13C chemical shifts
- Heavy atom relativistic effects
- XTB/DFT integration
Sources: GitHub repository, J. Chem. Theory Comput.
Key Strengths
Dual ML Correction:
- ΔcorrML: Accuracy correction to CCSD(T)
- ΔSO^ML: Relativistic correction for heavy atoms
- Combined: High accuracy + relativistic effects
- Systematic improvement
Heavy Atom Support:
- Spin-orbit corrections
- Relativistic effects on shifts
- Halogen and heavy element effects
- Beyond DFT accuracy
Grimme Lab Quality:
- Well-tested ML models
- XTB integration
- Consistent with Grimme ecosystem
- Published benchmarks
Inputs & Outputs
-
Input formats:
- DFT/XTB calculated shifts
- Molecular geometry
- ML model files
-
Output data types:
- Corrected chemical shifts
- ΔcorrML correction values
- ΔSO^ML correction values
- Final corrected shifts
Interfaces & Ecosystem
- XTB: Grimme's semi-empirical code
- DFT codes: Baseline shift calculation
- Python/PyTorch: ML backend
Performance Characteristics
- Speed: Fast (ML prediction)
- Accuracy: CCSD(T) + relativistic quality
- System size: Organic molecules
- Memory: Low
Computational Cost
- ML prediction: Seconds
- DFT/XTB baseline: Minutes (separate)
- Typical: Very efficient
Limitations & Known Constraints
- 1H and 13C only: No other nuclei
- Organic focus: Limited inorganic support
- Training dependent: Quality limited by training data
- XTB preferred: Best with Grimme ecosystem
Comparison with Other Codes
- vs CASCADE: ml4nmr has ΔSO^ML relativistic, CASCADE is Δ-ML accuracy
- vs ShiftML: ml4nmr is molecular, ShiftML is solid-state
- vs afnmr: ml4nmr is ML, afnmr is bio-NMR
- Unique strength: ML correction of NMR shifts with spin-orbit relativistic effects (ΔSO^ML)
Application Areas
Organic Chemistry:
- High-accuracy NMR prediction
- Heavy atom effects on shifts
- Structure verification
- Isomer identification
Organometallic Chemistry:
- Heavy atom relativistic effects
- Metal-containing molecules
- Halogen effects on shifts
- Spin-orbit coupling
Method Development:
- ML-NMR benchmarking
- Relativistic correction validation
- Chemical shift databases
- DFT functional assessment
Best Practices
DFT Baseline:
- Use consistent functional
- Include all atoms in calculation
- Use appropriate basis set
- Validate against experiment
ML Correction:
- Apply both ΔcorrML and ΔSO^ML
- Check domain of applicability
- Validate on test set
- Compare corrected vs uncorrected
Community and Support
- Open source on GitHub
- Developed by Grimme Lab (Bonn)
- Published in J. Chem. Theory Comput.
- Research code
Verification & Sources
Primary sources:
- GitHub: https://github.com/grimme-lab/ml4nmr
Confidence: VERIFIED
Verification status: ✅ VERIFIED
- Source code: ACCESSIBLE (GitHub)
- Documentation: Included in repository
- Published methodology: J. Chem. Theory Comput.
- Specialized strength: ML correction of NMR shifts with spin-orbit relativistic effects (ΔSO^ML)