Miroslav Lžičař
CTO & cofounder, Deep MedChem
I’m scaling intelligence into trillion-scale chemical spaces at Deep MedChem and exploring overlooked ideas in machine learning through Mereology Labs.
News
November 2026
September 2026
- OPDiv preprint: selecting high-scoring compounds under diversity constraints. Paper
Earlier updates
June 2026
- SynthonBench and Factorized Search and Cartography of Synthon-Based Chemical Spaces accepted to the ICML 2026 GenBio Workshop.
January 2026
- Our Deep MedChem team won the DYRK1B virtual screening challenge at 9ADD 2026. Post
November 2025
Current work

Work in progressResearch software, 2026
SynFactor
Distilling molecular property models into reusable synthon factors for fast approximate scoring. Run the example, inspect the evaluation, and see where the approximation breaks down.

ICML 2026 GenBio WorkshopChemical-space cartography, 2026
SynthonMap
Mapping combinatorial chemical spaces through factorized representations and product-mass density maps.

arXiv preprint, 2026Sep 2026
OPDiv: Optimal Selection of Top-K High-Scoring, Diverse Compounds
Lžičař, M.
Integer optimization for selecting high-scoring compounds under explicit diversity constraints. OPDiv measures the score–diversity tradeoff with fingerprint, shape, and electrostatic distances, providing a reference for evaluating virtual screening, molecular search, and generative models.
Selected results

ProjectMolecular search engine at Deep MedChem
CHEESE
From a custom billion-scale vector database and transformer I built before ChatGPT toward trillion-scale combinatorial search over shape and electrostatics.
Research conversations
I welcome conversations about research engineering, scientific ML, representations and evaluation. If a project connects to something you’re working on, email me.
For molecular-search workflows and commercial inquiries, contact Deep MedChem.


