Computational chemistry researcher and chemical engineering student building interpretable methods for non-covalent interactions, molecular screening, and scientific workflow automation.
I work at the intersection of supramolecular chemistry, quantum chemistry, chemical engineering, and scientific machine learning. My research focuses on representing non-covalent interactions in a form that is compact, physically meaningful, and useful for screening large chemical spaces.
The Kulkarni-NCI Fingerprint, or KNF, represents a non-covalent complex using nine geometric, electronic, and real-space interaction features. I developed KNF to connect molecular structure, interaction topology, and supramolecular stability without relying only on binding energy.
Publication: A Physics-Informed Fingerprint for Generalizable Prediction of Supramolecular Stability, The Journal of Physical Chemistry B.
K-UID is a scale-invariant address system for classifying non-covalent interaction environments from KNF features. It provides symbolic interaction identities that can be compared across molecular systems and length scales.
Publication: K-UID: A Scale-Invariant Topological Address System for Non-Covalent Interaction Classification Across Molecular Scales, Journal of Chemical Information and Modeling.
NCIForge is a computational platform for KNF, K-UID, reduced-density-gradient analysis, interaction screening, geometry processing, and backend-independent quantum-chemical workflows. Current development includes large-scale perturbation analysis, adaptive backend routing, and trust-controlled scientific machine learning.
I also work on deep eutectic solvents, CO2 capture, molecular host-guest screening, geometry initialization, and the connection between molecular-scale descriptors and chemical-engineering decisions.
| Project | Description |
|---|---|
| KNF-Predictor | Surrogate models for rapid prediction of KNF features and non-covalent interaction behaviour. |
| Geonit | Physics-inspired geometry initialization and conservative warm-start selection for molecular optimization. |
| Supramolecular-Stability-JCIM-Code | Reproducible scripts for molecular generation, xTB and DFT calculations, descriptor analysis, and machine learning. |
| DES-DESIGN | Computational workflow for deep eutectic solvent design and screening. |
| Interaction Heterogeneity Study | Statistical investigation of nonlinear relationships between interaction heterogeneity and supramolecular stability. |
| PFAS Interfacial Package | Computational tools for analysing PFAS behaviour at molecular interfaces. |
- Build descriptors that retain physical meaning.
- Separate predictive performance from chemical validity.
- Test whether conclusions remain stable across methods and datasets.
- Use machine learning to accelerate scientific workflows, not conceal uncertainty.
- Publish the code and evidence needed to reproduce reported results.
- Treat failed hypotheses and boundary cases as useful results.
Non-covalent interactions, conceptual density functional theory, scientific machine learning, molecular representation learning, supramolecular chemistry, deep eutectic solvents, CO2 capture, geometry optimization, and quantum-chemistry workflow engineering.
Languages and analysis: Python, NumPy, pandas, scikit-learn, PyTorch, PyTorch Geometric, RDKit, Pyomo, and scientific data visualization.
Computational chemistry: xTB, ORCA, Gaussian, Multiwfn, and real-space non-covalent interaction analysis.
Chemical engineering: Aspen Plus, process simulation, optimization, separations, and CO2-capture modelling.
I am interested in research collaborations involving interpretable molecular representations, non-covalent interactions, scientific software, conceptual DFT, molecular screening, and ML-assisted computational chemistry.
Contact: prasannakulkarni163@gmail.com