Applied AI research.
My research interests include applied machine learning, scientific machine learning, and the evaluation of AI systems. I am interested in how domain knowledge and reliable evaluation can make models more useful for real scientific and operational problems.
I co-authored a study on physics-informed machine learning for screening MOF/g-C3N4 heterojunction photocatalysts. I am looking to develop this research direction through MS/MPhil study and collaboration.
Co-authored publicationPhysics-informed machine learning screening and validation of metal-organic framework/g-C3N4 heterojunction photocatalysts.
- Authors, in order
- Abdullah Khan; Muhammad Saeed; Fakhrud Din; Farhat Ullah; Sami Ullah
- Journal record
- Chinese Journal of Physics · 103 (2026) · 1881-1895 · DOI 10.1016/j.cjph.2026.07.025
- Farhat Ullah's contribution
- Methodology, formal analysis, validation, visualization, writing the original draft, and review and editing.
The linked code repository is maintained under the first author's GitHub account; this page does not claim sole ownership of the study or repository.
Computational screening, not experimental confirmation.
The paper combines physics-informed descriptors with Random Forest and XGBoost rankings, then applies thermodynamic, pore-accessibility, and stability constraints. Internal evaluation and isolated external MOF families serve different purposes and should not be conflated.
- MOFs screened
- 20,152Integrated QMOF and CoRE-MOF dataset
- Internal ROC-AUC
- 0.912Ensemble result on 4,029 internal test structures
- Top-50 precision
- 98%Ranking result reported by the paper
- Final shortlist
- 983Computational candidates requiring further validation
The framework is scoped to closed-shell and main-group metal nodes. Open-shell 3d transition-metal systems require higher-fidelity modeling, and shortlisted materials still require prospective quantum-mechanical and experimental validation. Any compute reduction is a study-specific estimate against the paper's comparison, not a client cost-saving claim.
Research questions I want to develop.
- 01
How can domain knowledge improve machine learning when data or evaluation resources are limited?
- 02
How can evaluation identify unreliable model behavior before deployment?
- 03
How can research software make experiments easier to reproduce and compare?
Let's discuss what you're building.
Share the problem, the current setup, and what a useful outcome would look like. For research inquiries, include the topic or opportunity you would like to discuss.