Rajarshi
Sinha Roy
PhD Researcher in Informatics at the University of Leipzig, specializing in Artificial Intelligence for Drug Discovery under the supervision of Prof. Dr. Jens Meiler and Dr. Georg Künze.
My research integrates deep learning, bioinformatics, and structural biology to design interpretable computational models that predict enzyme function, protein stability, and complex structure-function relationships. Leveraging expertise in geometric deep learning, molecular dynamics simulation, and biomolecular modeling, my work bridges algorithmic innovation and molecular biophysics to enable predictive, explainable discovery of therapeutic enzymes and novel drug targets.
02. Ongoing Projects & Interests
Projects.
Computational mRNA Therapeutics
Collaborative Project with BioNTech
- • Utilized multiscale molecular dynamics (MD) simulations to analyze complex interactions between lipid nanoparticles (LNPs) and serum proteins.
- • Optimized computational design frameworks to enhance the delivery efficiency and stability of mRNA therapeutics.
ML-Assisted Directed Evolution
Plastic-Degrading Enzymes
- • Developed MLDE models to predict and optimize the correlation between catalytic activity and thermal stability for PHL7 enzyme variants.
- • Engineered fitness functions that improved the alignment between in silico sequence predictions and experimental wet-lab outcomes.
Flow Matching & Backmapping
Deep Learning for Structural Biology
- • Developed flow-matching-based deep learning methods to backmap coarse-grained (CG) structures to all-atom (AA) resolutions.
- • Enabled high-fidelity structural reconstruction across diverse biomolecules, including lipids, proteins, DNA, and sugars.
MD Surrogate Modeling
Physics-Aware Architecture Search
- • Designed a staged, physics-aware architecture search and optimization framework to serve as a surrogate model for molecular dynamics.
- • Accelerated simulation workflows by embedding physical constraints directly into neural network architectures.
SE(3) Diffusion World Models
Protein Conformational Dynamics
- • Engineered a reference-conditioned SE(3) diffusion world model to learn and predict local protein conformational dynamics.
- • Generated accurate rollouts of complex structural transitions directly from raw molecular dynamics trajectories.
Deeplearning-Based Protein Design
pH-Dependent Sequence Generation
- • Designed an Equivariant Graph Neural Network (EGNN) pipeline to generate de novo protein sequences conditioned on specific pH optimum requirements.
- • Enabled conditional sequence generation for tailoring therapeutic enzymes and biocatalysts to precise industrial and physiological pH microenvironments.
03. Experience & Education
Trajectory.
PhD Researcher
University of Leipzig, Germany
Collaborative project with BioNTech. Developing SE(3)-equivariant GNNs for enzyme pH-optima prediction and designing MPNN-Transformer architectures.
M.Sc. Computer Science
St. Xavier's College, Kolkata
CGPA: 8.36. Dissertation on Optimization in Image Processing and Deep Learning.
Academic Researcher
Bioinformatics and AI Lab
Designed hybrid GAN + DCNN architectures for MRI-based Alzheimer's progression prediction. Built robust preprocessing evaluation pipelines.
B.Sc. Computer Science
GGDC, Singur
CGPA: 8.83. Core focus on Artificial Intelligence, Database Systems, and Applied Computing.
04. Selected Works
Publications.
Predicting Enzyme pH Optima from Structure Using Equivariant Graph Neural Networks
R. SinhaRoy, C. Clauss, I. Ivanikov, and G. Künze • bioRxiv (2026)
A Hybrid Deep Learning Framework to Predict Alzheimer's Disease Progression Using GAN and DCNN
R. Sinha Roy and A. Sen • Arabian Journal for Science and Engineering (2023)
pHoptNN: Predicting Enzyme pH-Optima from Structure Using Equivariant Graph Neural Networks
R. SinhaRoy • Oral Presentation, EUROSETTACON (2025)
Overlapped Fingerprint Separation Using Graph-Based Model
S. Sen, S. Sonali, R. SinhaRoy et al. • IEEE SILCON (2022)
05. Capabilities
Technical Skills.
Deep Learning & Generative AI
Transformers, Diffusion Models, SE(3) World Models, Flow Matching, Agentic AI Architectures, State Space Models (SSMs / Mamba), Graph Neural Networks (GNNs).
AI Frameworks & Libraries
PyTorch, TensorFlow, Jax, PyTorch Geometric, Hugging Face.
Structural Biology & AI Tools
AlphaFold (2/3), RFDiffusion, ProteinMPNN, ESMFold, ChimeraX.
MD & Physics Simulation
GROMACS, Martini Coarse-Grained Force Field, MDAnalysis, MDTraj, BioPython, All-Atom & Coarse-Grained Backmapping.
Languages & Core Tech
Python, C/C++, SQL, MATLAB, Bash, Git / CI-CD.
Domain Expertise
- • Computer Science & AI: Geometric Deep Learning, Generative AI Architectures, Physics-Aware AI Search, Equivariant Graph Networks, Data Mining & Representation Learning.
- • Biophysics & Computational Biology: Structural Bioinformatics, Molecular Dynamics (MD) Surrogate Modeling, De Novo Protein Design, Enzyme Kinetics & Directed Evolution.