Excited to share our new work with @Cellibre1 on machine learning-guided enzyme engineering for cannabinoid biosynthesis. 🌱🧬 We engineered olivetolic acid cyclase (OAC) to improve pathway flux and selectively produce major and minor cannabinoids in yeast. A short thread 👇
Users praise pathway-level machine learning for engineering difficult enzymes in yeast-based cannabinoid biosynthesis because it can redirect metabolic flux and expand access to major, minor, and new compounds.
No Digg Deeper questions have been answered for this story yet.
First, we built an end-to-end cannabinoid pathway in Yarrowia lipolytica.
It produced OA, CBGA, and THCA but accumulated PDAL and OL, revealing OAC as a bottleneck that must outcompete spontaneous side reactions.
Traditional engineering fell short: homologs,recombination & mutagenesis yielded limited gains. So we learned the sequence–function landscape directly via: •Multi-task NNs of pathway-level product & byproduct measurements •VAE of learned evolutionary constraints from homologs
I am excited to share our new preprint with @Cellibre1 on machine learning-guided enzyme engineering for cannabinoid biosynthesis!
Preprint: https://www.biorxiv.org/content/10.64898/2026.06.12.731972v1 Code: http://github.com/RomeroLab/ml-guided-cannabinoid-biosynthesis
Excited to share our new work with @Cellibre1 on machine learning-guided enzyme engineering for cannabinoid biosynthesis. 🌱🧬 We engineered olivetolic acid cyclase (OAC) to improve pathway flux and selectively produce major and minor cannabinoids in yeast. A short thread 👇
This work shows how pathway-level ML can engineer difficult enzymes, redirect metabolic flux, and expand access to major, minor, and potentially new-to-nature cannabinoids.
A wonderful collaboration between Cellibre and the Romero Lab member @NathanielBlalo2 !
Paper + code 👇
In the full pathway, CsOAC1.295 enabled 108 mg/L THCA, while CsOAC1.285 enabled 78 mg/L THCVA. These titers were 13.5× and 16.25× higher than prior engineered yeast systems and the first reported in Y. lipolytica to our knowledge.
Using multi-objective design, we balanced predicted function with evolutionary plausibility across three design–build–test–learn rounds. • CsOAC1.295: +300 mg/L OA • CsOAC1.285: +91.5 mg/L DVA
📄https://www.biorxiv.org/content/10.64898/2026.06.12.731972v1.article-metrics
💻 http://github.com/RomeroLab/ml-guided-cannabinoid-biosynthesis
🏢 http://cellibre.com
#MachineLearning #SyntheticBiology #MetabolicEngineering #ProteinEngineering
