1
Genome & annotation
Curate genomic data for your cell line — most industrially relevant lines already have sequenced genomes and public models.
Technology
Genome-scale models meet hybrid AI
A five-stage pipeline rooted in a decade of systems biotechnology at NOVA University Lisbon — validated on CHO, HEK, and other mammalian platforms.
The pipeline
From public genomes to optimized formulations — with a fraction of the experiments traditional screening requires.
2
Genome-scale model
Build or adapt a GEM: metabolites, reactions, and transport tailored to your medium and cell line.
3
Designed experiments
A targeted design of experiment yields the exchange fluxes and omics readouts needed to train hybrid models — not exhaustive screening.
4
Hybrid semi-parametric AI
Blend mechanistic FBA with data-driven constraints (PCA / ML) so predictions respect both biology and your process data.
5
Optimize & deliver
Multi-objective optimization proposes media, feed composition, and feeding strategy — then we validate in the lab.
- 1
Genome & annotation
Curate genomic data for your cell line — most industrially relevant lines already have sequenced genomes and public models.
- 2
Genome-scale model
Build or adapt a GEM: metabolites, reactions, and transport tailored to your medium and cell line.
- 3
Designed experiments
A targeted design of experiment yields the exchange fluxes and omics readouts needed to train hybrid models — not exhaustive screening.
- 4
Hybrid semi-parametric AI
Blend mechanistic FBA with data-driven constraints (PCA / ML) so predictions respect both biology and your process data.
- 5
Optimize & deliver
Multi-objective optimization proposes media, feed composition, and feeding strategy — then we validate in the lab.
At a glance
Mechanistic models + machine learning
Genome-scale metabolic networks provide scientific structure. Hybrid semi-parametric models learn from your data. Optimization searches the space of media and feeds for your objectives.
Mechanistic
GEM stoichiometry, mass balance, pathway constraints
Data-driven
PCA / ML constraints from targeted experiments
Optimization
Multi-objective media, feeds & feeding strategy
Research foundation
Published science behind the platform
Levacells builds on peer-reviewed work by Rui Oliveira’s group at LAQV REQUIMTE (NOVA University Lisbon) and João Ramos’s hybrid genome-scale modelling — including culture media design, dynamic cell models, and functional enviromics.
- 2022Bioprocess and Biosystems Engineering
Genome-scale modeling of Chinese hamster ovary cells by hybrid semi-parametric flux balance analysis
João R. C. Ramos, Gil P. Oliveira, Patrick Dumas, Rui Oliveira
HybridFBA for CHO culture media design — core Levacells methodology.
- 2023Frontiers in Bioengineering and Biotechnology
Hybrid deep modeling of a CHO-K1 fed-batch process: Combining first principles with LSTM neural networks
José Pinto, João R. C. Ramos, Rafael S. Costa, Rui Oliveira
Deep hybrid models linking GEM-scale metabolism to fed-batch dynamics.
- 2020Biotechnology and Bioengineering
A dynamic model linking cell growth to intracellular metabolism and extracellular by-product accumulation
João R. C. Ramos, A. G. Rath, Y. Genzel, V. Sandig, U. Reichl
Structured dynamic models for mammalian cell culture optimization.
- 2014Biotechnology and Bioengineering
Culture media engineering by projection to latent pathways: The case of Pichia pastoris
Rui Oliveira, Inês A. Isidro, Rui M. C. Portela
Latent-pathway projection for rational media formulation.
- 2010Patent / MediaOmics spin-off technology
A cell functional enviromics method for the engineering of cell culture media (patent family)
Rui Oliveira, João M. L. Dias, Ana Raquel S. Ferreira
Functional enviromics — joint screening of cell functions and medium factors.
Ready to optimize your cell culture?
Tell us about your cell line and process goals — we will explore whether our approach fits your timeline.