MSc in Computational Sciences and Engineering
- Presidential Fellowship for MSc entrance
- Thesis direction: generative models for peptide and binder design
- Advised by Prof. Attila Gürsoy and Prof. Özlem Keskin
Curriculum vitae
noroozi.alirezaa@gmail.com · Persian (native), English (fluent)
Education
BSc in Computer Engineering
- GPA: 17.32/20 (3.46/4.0)
- Focus on machine learning and computational biology
- Thesis on paratope prediction using protein language models, published in Scientific Reports
- Ranked 31st among 150,000+ participants in Iran’s national university entrance exam (2020)
Research experience
Research Intern — peptide binder design, molecular dynamics
- Supervised by Hamed Khakzad and Yasaman Karami
- Developed IDPepDesign, a target-conditioned insertion–deletion diffusion model for peptide binder design, built on SCISOR with pooled-target conditioning and residue-level cross-attention
- Contributed to BinderEval, which evaluates designed binders using molecular dynamics data and COMPASS rather than single-structure confidence scores
- Released a new version of COMPASS
- Gave a talk at PhD Pizza Time, Inria Nancy (Sep 2026)
Graduate Researcher — antimicrobial peptide design
- AMP-DiT: a classifier-conditioned discrete diffusion transformer that generates antimicrobial peptides directly in sequence space, avoiding the representation mismatch of protein language models trained on full-length proteins. Accepted as a poster at the GenBio Workshop at ICML 2026 and submitted to the AMP Challenge
- AMP-BindGen: guided design of peptide binders and inhibitors against the carbapenemases NDM-5 and KPC-3, with a composite ranking scheme over interface confidence, catalytic-site engagement, predicted activity and toxicity. Designs have been validated experimentally in the lab
- TED: an end-to-end sequence-to-sequence transformer predicting protein domain boundaries and CATH identifiers directly from sequence, trained on ~186M sequences with LoRA fine-tuning over ESM-2 embeddings
- BoltzGen-ADC: generative design of protease-cleavable peptide linkers for antibody–drug conjugates, submitted to the Merck Challenge (with Mahda Ghasemiagreh)
- Collaboration with KUISCID and the School of Medicine on carbapenem resistance
Research Assistant — transfer learning for paratope prediction
- Bachelor project under the supervision of Dr. Koohi
- ParaAntiProt: sequence-only paratope prediction via transfer learning from antibody and protein language models
- Published in Scientific Reports (2024)
Research Intern — federated learning for medical data
- Summer internship under the supervision of Dr. Kersten-Oertel
- Compared and implemented federated learning algorithms for imputing Parkinson’s disease clinical assessments; published in IEEE JBHI (2025)
- Implemented PW, FedAvg and FedOpt in PyTorch, including work on a convergence proof for PW
- Developed federated algorithms for non-identically distributed data, and GAN-based methods for prediction from distributed data
Research Assistant — attention models for binding affinity
- MHC–peptide binding prediction using attention mechanisms, supervised by Dr. Koohi
Publications
Preprints and submissions
IDPepDesign: Insertion–Deletion is all peptide design needs
Preprint · ML4SM Workshop, NeurIPS 2026
BinderEval: evaluating designed protein binders with dynamics data and COMPASS
Preprint
BoltzGen-ADC: generative design of universal protease-cleavable peptide linkers for antibody–drug conjugates
Merck Challenge submission
Peer-reviewed
AMP-DiT: Antimicrobial Peptide Design with AMP-classifier Conditional Diffusion Transformers
GenBio Workshop at ICML 2026 (poster)
Bridging the Gaps: Imputation of Parkinson’s Disease Clinical Assessments With Federated Learning
IEEE Journal of Biomedical and Health Informatics
ParaAntiProt provides paratope prediction using antibody and protein language models
Scientific Reports 14, 29141
Talks and posters
- Talk, PhD Pizza Time, Inria Nancy, Sep 2026
- Posters, ECCB 2026, Geneva — AMP-BindGen and AMP-DiT
- Posters, Koç University AI Symposium, Apr 2026 — AMP design and TED; joint talk on carbapenem resistance and AMP design with AI
Industry experience
AI Engineer in Computational Biology
- Deep learning models for protein design
- Taxonomy lineage generation: a transformer that generates full taxonomy lineages from protein sequences; designed the training pipeline, tokenisation and loss, and addressed token collapse with curriculum learning and embedding regularisation
- TED domain classifier: multi-class classification of immunoglobulin domains with ProtT5 and ESM-2, including large-scale preprocessing and GPU-cluster fine-tuning (code)
- IgG1 binder design: generative protein models with structural constraints and PLM embeddings to guide binder specificity and stability
- Built an AI chatbot backend using LLM tool calling
Back-end and web developer (PHP)
- Built a stock-trading website using the Laminas framework
- Developed an Excel and PDF report generator
Machine Learning Intern
- Summer internship on machine learning for medical imaging
Honors and awards
- 2025 Presidential Fellowship for MSc entrance, Koç University
- 2024 Vector Scholarship for MSc entrance, York University (17,500 CAD; deferred)
- 2020 Ranked 31st of 150,000+ in Iran’s national university entrance exam
- 2019 Silver Medal, Iran’s National Mathematical Olympiad
- 2018 Bronze Medal, Iran’s National Mathematical Olympiad
Service and teaching
Reviewing
- GenBio Workshop (Generative and Agentic AI for Biology), ICML 2026
- ML4SM Workshop, NeurIPS 2026
Teaching assistance
- Python Programming, Koç University
- Head TA, Probability and Statistics (2 semesters), Sharif University
- Head TA, Compiler Design, Sharif University
- Project lead TA, Introduction to Bioinformatics, Sharif University
- Discrete Structures, Sharif University
- Mathematics Olympiad tutor, Salam High School (2021–2023)
Research interests
Design
- Peptide and protein binder design
- Generative models for biological sequences
- Discrete and insertion–deletion diffusion
- Antibody design and paratope prediction
Evaluation and modelling
- Molecular dynamics for flexibility and binding
- Evaluation beyond static confidence scores
- Protein–ligand cofolding and adaptive resolution
- Protein language models and transfer learning
Technical skills
- Languages: Python, C++, JavaScript, PHP
- ML: PyTorch, Hugging Face, scikit-learn, distributed and multi-GPU training
- Structural biology: Boltz, BoltzGen, AlphaFold-class predictors, RFdiffusion, ESM-2/ESM-C, ProtT5, Biopython, GROMACS-style MD analysis
- Other: Git, Docker, Linux, SLURM clusters
Languages
- Persian: native
- English: fluent — TOEFL 97 (R 26, L 25, S 25, W 21)