Hosein Fooladi

Machine Learning Researcher

CV

Download Complete CV

Get the full PDF version with detailed experience and achievements

Download Full CV (PDF)

Last updated: October 2026

Profile

Machine learning researcher focused on models that hold up on data they have never seen. One thread runs through eleven years of work: from mechanical engineering and robotics, through mathematical models of stem-cell self-organization, to leading industrial ML teams in drug discovery, and now a PhD on domain generalization in chemical space. I care about evaluation that mirrors the intended application, reproducible pipelines, and tools other people can rerun and trust.

  • 5 papers since 2023 (three in JCIM, one in AI in the Life Sciences, one on ChemRxiv)
  • 12+ open-source packages, including Rust tooling for structural biology and cheminformatics and the GNNs-for-Chemists course (180+ GitHub stars)
  • Led teams of 5+ ML engineers and computational chemists in two biotech companies

Education

Ph.D. in Pharmaceutical Sciences (Cheminformatics)

University of Vienna | Vienna, Austria | 2022 - Present
Kirchmair lab (Comp3D), Christian Doppler Laboratory for Molecular Informatics in the Biosciences (CD-Lab MIB), with industry partners Boehringer Ingelheim and BASF
Thesis: Machine learning models for domain generalization in chemical space

M.Sc. in Biomedical Engineering

Sharif University of Technology | Tehran, Iran | 2015 - 2017
Thesis: A multicellular model of stem-cell self-organization, extending the Waddington landscape with cell-cell communication (published in Bioinformatics, 2019)
🏆 Best Master’s Student Award

B.Sc. in Mechanical Engineering

Amirkabir University of Technology (Tehran Polytechnic) | Tehran, Iran | 2009 - 2014
Focus on dynamics and robotics; design and experimental study of a passive walking biped

Work Experience

Head of Machine Learning, then Chief Data Scientist

Celeris Therapeutics | Graz, Austria | Feb 2021 - Feb 2022
Head of Machine Learning from Feb 2021; Chief Data Scientist from Dec 2021

  • Led a team of 5+ ML engineers and computational chemists working on targeted protein degradation
  • Owned the ML platform end to end: repository standards, CI on Azure Pipelines, Docker CPU/GPU images, AWS compute (EC2 spot fleets, ParallelCluster with Slurm, S3), experiment configuration and tracking
  • Workstreams: virtual screening, protein-protein interface prediction, PROTAC linker generation and ternary complex prediction
  • Co-developed BOTCP, a Bayesian-optimization method for PROTAC ternary complexes: constrained conformer generation, combined fitness scoring and cluster deployment; a near-native cluster ranked in the top 15 for 16 of 22 benchmark complexes, in under two hours per complex on 128 CPU cores

Freelance Machine Learning Engineer

Vetevo | Berlin, Germany (remote) | 2022, four months

  • Built a parasite-egg detection service for veterinary microscopy images end to end, alone
  • Labelbox-to-YOLO dataset tooling (1,955 images, 3,132 boxes, 10 classes with 15x class imbalance), YOLOv5m training at 640 px, and a Flask inference endpoint
  • Delivered mAP@0.5 of 0.92 and recall of 0.91 within four weeks

Senior Data Scientist - Cheminformatics/ML Expert

AI VIVO | Cambridge, UK | Apr 2019 - Dec 2020
Part of a 5+ team of ML, biology and chemistry specialists working on a single on-premise GPU node

  • “Virtual cells”: predicted small-molecule perturbation responses across cell lines on LINCS L1000 (978 landmark genes, 1.3M profiles, ~20k compounds) with a Dr.VAE-family latent-transition model (overview with figures); open-sourced the data processing as lincs_processing
  • Drug repositioning via transfer learning from ChEMBL (2M+ compounds) to rare-disease targets
  • VAE-based de novo molecular design with multi-objective optimization
  • Drug-combination synergy prediction (O’Neil, NCI-ALMANAC, DrugComb, DrugCombDB)
  • Reproducible Nextflow workflows with cold-compound and cold-cell-line splits and explicit baselines; PyTorch, TensorFlow, Weights & Biases, FastAPI and Gradio

Chief Scientific Officer (CSO)

Shenakht Pajouh | Tehran, Iran | May 2018 - Dec 2019

  • Integrated psychological knowledge with machine learning for automated mental health assistance
  • Led scientific strategy and research development

Machine Learning Researcher

Cambridge Systems Biology Centre | Cambridge, UK | Feb 2017 - Jan 2018

  • Deep learning on single-cell RNA-seq data: autoencoders and cell-type classifiers
  • First models on high-dimensional expression data after mechanistic ODE modelling in the master’s thesis

Bioinformatics Researcher

Royan Institute | Tehran, Iran | Jan 2017 - Aug 2017

  • Reconstructed context-specific metabolic networks from gene expression data
  • Applied computational methods to systems biology problems

Teaching Assistant

Sharif University of Technology | Tehran, Iran | Spring 2017

  • Advanced Bioinformatics course
  • Systems Biology course

Technical Skills

Programming Languages

  • Python (Advanced): daily driver for research and production code (NumPy, Pandas, SciPy, pydantic)
  • Rust (Intermediate): high-performance parsers and tokenizers with Python bindings (pdbrust, sdfrust, rustmolbpe)
  • R (Advanced): statistical modelling, Bioconductor
  • C++ (Intermediate): performance-critical implementations

Machine Learning

  • Deep Learning: PyTorch, JAX, TensorFlow; graph neural networks, transformers, VAEs and diffusion models
  • Classical ML: scikit-learn, XGBoost, LightGBM
  • Methods: out-of-distribution evaluation, transfer and meta-learning, Bayesian optimization and active learning, causal inference

MLOps & Infrastructure

  • Experimentation: Hydra configs, MLflow, Weights & Biases
  • Pipelines & compute: Nextflow, Slurm / AWS ParallelCluster, Docker CPU/GPU images, AWS (EC2, S3)
  • CI/CD & serving: GitHub Actions, Azure Pipelines, FastAPI, Gradio

Computational Chemistry & Bioinformatics

  • Cheminformatics: RDKit, DeepChem, Open Babel, ChEMBL and FS-Mol data pipelines
  • Structure-based modelling: protein-ligand docking, ML-based pose sampling (DiffDock-L), AlphaFold2-Multimer, molecular dynamics
  • Bioinformatics: single-cell RNA-seq, perturbation data (LINCS L1000), network and systems biology

Research Expertise

  • Out-of-Distribution Generalization: defining and evaluating OOD data in chemical space; 14 models x 8 datasets x 10 splitting strategies (ALineMol)
  • Transfer and Meta-Learning: quantifying bioactivity task hardness and task relations to guide source-task selection (THEMAP)
  • Targeted Protein Degradation: PROTAC design and ternary complex prediction with Bayesian optimization
  • Perturbation-Response Modelling: “virtual cell” models of transcriptional responses to small molecules
  • Systems Biology: mechanistic models of stem-cell self-organization and pattern formation

Publications

Peer Review Activities

Talks

Teaching

Open Source Software

Research Libraries

  • THEMAP: task hardness estimation for molecular activity prediction (on PyPI)
  • ALineMol: evaluating ML models on out-of-distribution data in the chemical domain
  • lincs_processing: parsing and processing of the LINCS L1000 dataset
  • molax: molecular active learning in JAX
  • bayesoptimol: Bayesian optimization and active learning for drug discovery
  • chembl-pdb-linker: linking ChEMBL bioactivity data with PDB structures

Rust Tooling

  • pdbrust: PDB parser, 40-260x faster than pure Python
  • sdfrust: SDF/MOL2 parser at roughly 220k molecules per second
  • rustmolbpe: BPE tokenizer for SMILES with Python bindings
  • rustdock-vina: AutoDock Vina in Rust (work in progress)

Course Repositories