Lugano, Switzerland
I take drug discovery programs from computational design to the clinic. Over the past 15+ years, I have designed and shipped production cheminformatics software, physics-based simulations, and machine learning systems in C, C++, CUDA, and Python.
The computational discovery pipelines I led at Endogena Therapeutics helped advance one small-molecule candidate into human clinical trials and a second to IND approval. At Chemify, I lead the computational architecture across 6+ commercial discovery programs for global pharma and biotech partners, combining trillion-scale chemical space search, biomolecular cofolding, and free-energy methods.
- From In Silico to Human Clinical Trials (Endogena Therapeutics): Led the machine learning and structure-based design pipelines behind 1 small-molecule candidate in human clinical trials and a 2nd compound advanced to IND, co-authoring core composition-of-matter patents.
- Trillion-Scale Chemical Space Search (Chemify Ltd.): Built de novo generation and synthon-space indexing algorithms that query multi-trillion virtual compound spaces in under 180 ms to drive automated robotic synthesis for the Gates Foundation, Dewpoint Therapeutics, and confidential biopharma partners.
- Commercial Pipeline Delivery (Chemify Ltd.): Led computational design across 6+ target portfolios with a 100% project progression rate (zero early terminations), delivering potency-enhanced lead series and building foundational computational tools that supported a new automated synthesis facility and a $50M Series B financing.
- Core Open-Source Machine Learning (CatBoost / Hoffmann-La Roche): Authored native multi-task regression with missing-value support in Yandex's CatBoost (PR #1680), solving the sparse experimental bioassay matrix problem common in industrial screening.
-
Staff Computational Scientist | Chemify Ltd. (Glasgow, UK & Lugano, Switzerland) |
2023 - Present
Lead computational architect for trillion-scale de novo molecular generation, biomolecular cofolding (Boltz-2 fine-tuning), autonomous agentic DMTA workflows, and multi-parameter lead optimization across commercial programs. -
Head of AI and Cheminformatics | Endogena Therapeutics (Zurich, Switzerland) |
2021 - 2023
Directed AI, QSAR/QSPR, and structure-based discovery for regenerative medicine and fibrosis; architected the company's multi-GPU compute infrastructure and scientific data platforms. -
Postdoctoral Research Scientist | Hoffmann-La Roche (Basel, Switzerland) |
2019 - 2021
Developed multi-task deep learning and gradient boosting methods for on-target potency and off-target safety profiling across sparse screening matrices. -
Research Scientist | Dalle Molle Institute for Artificial Intelligence (IDSIA) (Lugano, Switzerland) |
2017 - 2019
Pioneered 3D geometric deep learning on voxelized quantum and electrostatic molecular fields for early-stage drug discovery (Innosuisse ARCADIA project). -
Postdoctoral Researcher & External CADD Consultant | University of Geneva & Firmenich (Geneva, Switzerland) |
2014 - 2017
Independently discovered novel LIM Kinase inhibitors via structure-based modeling; built physicochemical property prediction platforms with the Swiss Institute of Bioinformatics (Vital-IT).
- Structure, Physics & Target Modalities: Boltz-2 and AlphaFold cofolding, Relative Binding Free Energy (OpenFE RBFE, AMBER-TI), Molecular Dynamics, QM/MM, docking (GOLD, AutoDock, DiffDock). Applied across Ion Channels, Protein-Protein Interactions (PPIs), Peptidomimetics, Small Molecules, Kinases, HDACs, Sirtuins, and Epigenetics (Oncology, Regenerative Medicine, Fibrosis/IPF, CNS, Infectious Disease).
- Cheminformatics, Machine Learning & Agentic AI: Production cheminformatics (RDKit, synthon indexing, QSAR/QSPR, NIPALS PCA, PLS, Consensus PCA), Multi-Task Learning, 3D Voxel CNNs, GNNs, PyTorch, CatBoost, and autonomous LLM/agentic workflows (RAG and multi-agent tool orchestration connecting literature/patent data to simulation pipelines).
- High-Performance Engineering & MLOps: Modern C++ (C++20), ANSI C (C99), CUDA, Python (ctypes/C-API/FFI), Go, Rust; Apache Airflow, Dagster, MLflow, Kubernetes, Docker, and multi-GPU Linux HPC clusters.
|
|
libscientific: Zero-Dependency C Numerical Engine High-performance C99 library for multivariate statistical analysis, numerical optimization, and chemometrics with full Python FFI bindings. Designed for speed, reproducibility, and zero external dependencies. -blue?style=flat-square){kind=icon}
|
|
|
CatBoost: Multi-Task Learning on Sparse Bioassay Matrices Engineered core C++/Python support in Yandex's CatBoost for multi-task regression with missing target labels, enabling simultaneous modeling of correlated biological endpoints across incomplete screening matrices. -purple?style=flat-square){kind=icon}
|
|
|
QStudioMetrics: C++20 Multivariate & QSAR Suite Cross-platform C++20 desktop application for exploratory data analysis, PCA/PLS, MLR, and rigorous QSAR model validation (K-Fold cross-validation, Leave-One-Out, and Y-Scrambling). 
|
|
|
MolecularFramework: 3D Conformer & Electrostatic Field Engine C/C++ molecular modeling library implementing Iterative Closest Point (ICP) 3D conformer alignment, van der Waals surface shape similarity, and 3D molecular electrostatic potential field calculation. 
|
|
|
DeepMolecularNetwork: 3D Quantum & Electrostatic Voxel CNNs Multi-modal deep learning framework utilizing 3D voxelized grids of quantum/force-field electrostatic potentials (B3LYP CHELPG, MMFF94, EEM) and electron density for physicochemical property prediction. -b31b1b?style=flat-square){kind=icon}
|
|
|
DynaStI & PyLSS: Physicochemical Property & Retention Modeling Predictive physicochemical modeling platform deployed at the Swiss Institute of Bioinformatics (Vital-IT) alongside PyLSS (Linear Solvent Strength simulation) for automated structure annotation and biomarker discovery. 
|
- libscientific: A Powerful C Library for Multivariate Analysis
G. M. Randazzo - The Journal of Open Source Software (JOSS), 8(87), 5420 (2023). DOI: 10.21105/joss.05420 - Steroid identification via deep learning retention time predictions and two-dimensional gas chromatography-high resolution mass spectrometry
G. M. Randazzo, A. Bileck, A. Danani, B. Vogt, M. Groessl - Journal of Chromatography A, 1612, 460661 (2020). ScienceDirect / DOI: 10.1016/j.chroma.2019.460661 - DynaStI: A Dynamic Retention Time Database for Steroidomics
S. Codesido, G. M. Randazzo, F. Lehmann, et al. - Metabolites, 9(12), 291 (2019). DOI: 10.3390/metabo9120291 - A Rational Approach for the Identification of Non-Hydroxamate HDAC6-Selective Inhibitors
L. Goracci, N. Deschamps, G. M. Randazzo, et al. - Scientific Reports, 6, 29086 (2016). DOI: 10.1038/srep29086 - Discovery of Novel, Potent, and Specific Cell-Death Inducers in the Jurkat Acute Lymphoblastic Leukemia Cell Line
E. Carosati, N. van den Hofel, M. Reif, G. M. Randazzo, et al. - ChemMedChem, 10(9), 1541-1554 (2015). DOI: 10.1002/cmdc.201500222
Complete list of 20+ peer-reviewed articles available on Google Scholar.
- PhD in Chemical Science | University of Perugia, Italy (2010 - 2013)
Quantitative Structure-Property/Retention Relationships, physicochemical modeling, and metabolomics. - Master of Science in Organic Chemistry | University of Perugia, Italy (2008 - 2010)
Computational selection and organic synthesis of biologically active compounds. - Bachelor of Science in Chemistry | University of Catania, Italy (2004 - 2008)
Microwave-assisted organic synthesis of therapeutic candidates.
I have been writing scientific and cheminformatics software from first principles since 2008, long before modern AI assistants, and today I combine that mathematical and systems foundation with generative and agentic tools to move faster at scale.
At heart, I am a lifelong Linux/Unix enthusiast (Gentoo, Debian, Fedora with KDE Plasma) and open-source builder. Outside of drug discovery, my projects include CheapSecurity (self-hosted CCTV for Linux single-board computers, 220+ stars), DeepMatrixInversion, TOO (TensorFlow on Odroid), upstream Linux Kernel patches, and Mindguard.
When I am away from a terminal, I am an avid distance runner with a 1h 14m Half Marathon (21.1 km) personal best (70th of 10,806 overall; 9th of 640 in SM40), and I enjoy spending time with my family, playing the French horn, inline skating, and restoring vintage computers.
Always happy to connect with fellow scientists, founders, and R&D leaders on AI-driven drug discovery, biomolecular simulation, and high-performance cheminformatics.
- Email: gmrandazzo@gmail.com
- LinkedIn: linkedin.com/in/gmrandazzo
- GitHub: github.com/gmrandazzo
- Google Scholar: Giuseppe Marco Randazzo
- Location: Lugano, Switzerland