Multi-Modal Foundation · openadmet
CYP Inhibition Model
A multi-task baseline model predicting pIC50 inhibition values across four cytochrome P450 isoforms.
What CYP Inhibition Model is
The CYP Inhibition Model is an ADMET prediction baseline developed by openadmet. It predicts the inhibition potency of chemical compounds against four major cytochrome P450 enzyme isoforms: CYP1A2, CYP2D6, CYP3A4, and CYP3C9. Cytochrome P450 enzymes play a central role in drug metabolism, making the assessment of CYP inhibition critical in preclinical drug discovery to identify potential metabolic liabilities and adverse drug–drug interactions.
The model outputs predicted pIC50 values (the negative logarithm of the half-maximal inhibitory concentration, IC50) simultaneously for each of the four CYP targets using canonical SMILES strings as chemical input.
This system serves as a publicly available baseline implementation. Its developers report that overall model performance is poor, meaning it is intended primarily as an initial benchmark rather than a reliable predictive tool for production drug design pipelines.
How it works
The model employs the multitask CheMeleon architecture to jointly evaluate inhibition against four enzymatic targets: CYP1A2, CYP2D6, CYP3A4, and CYP3C9.
It was trained on curated pIC50 bioactivity data obtained from ChEMBL. The training configuration is a no-split setup, meaning a training fraction of 1.0 was utilised with no data allocated to validation or testing sets.
Key facts
- Model architecture
- Multitask CheMeleon
- Model category
- ADMET Prediction
- Targets
- CYP1A2, CYP2D6, CYP3A4, CYP3C9
- Target metric
- pIC50
- Training dataset
- ChEMBL curated data
- Dataset split
- No split (training fraction 1.0; no validation or test set)
- Release date
- 13 November 2025
What you can use it for
- Multi-target CYP inhibition benchmarking
- Evaluating baseline predictive accuracy across CYP1A2, CYP2D6, CYP3A4, and CYP3C9 when developing novel ADMET modeling architectures.
- Preliminary small-molecule screening
- Generating initial pIC50 estimates from SMILES strings across four primary cytochrome P450 enzymes during exploratory drug discovery.
- Pipeline testing with the Anvil framework
- Validating multi-task inference scripts and input–output workflows inside the openadmet ecosystem before deploying higher-performing models.
Run it here
We list this model but cannot run it yet: no pipeline_tag — needs a human.
Everything on this page still applies. Running it needs a loader for this repository, not a bigger machine.
Strengths and limits
Good at
- Predicts inhibition across four distinct cytochrome P450 isoforms within a single multi-task CheMeleon run.
- Accepts standard canonical SMILES representations to output quantitative pIC50 values.
- Integrates into the openadmet Anvil framework and containerised Docker workflows for standardized execution.
Watch out for
- The developers report that model performance is poor, advising researchers to proceed with caution.
- Trained without separate validation or test splits (1.0 training fraction), precluding standard internal validation benchmarks during training.
- Does not generate uncertainty estimates or standard deviations because it is a single model rather than an ensemble.
Inputs and outputs
You provide
Sequence or text
A protein or nucleotide sequence, or free text — whichever this model was trained on.
You get back
Mean-pooled representation, one vector per input.
Licence
Check the licence before commercial use · apache-2.0
The apache-2.0 licence carries terms worth reading before you depend on it.
Questions
Can I use CYP Inhibition Model commercially?
Check the licence before commercial use. The apache-2.0 licence carries terms worth reading before you depend on it.
What input format does the model require?
The model takes molecular structures formatted as canonical SMILES strings, provided via a tabular file such as a CSV.
Which enzyme targets does the model cover?
It predicts inhibition for four cytochrome P450 isoforms: CYP1A2, CYP2D6, CYP3A4, and CYP3C9.
Why are the standard deviation columns empty in the output?
The prediction pipeline only calculates standard deviations (uncertainty estimates) when running an ensemble of models, which this single baseline model does not include.
Can this model be used reliably for lead optimisation decisions?
The developers state that the model performance is quite poor and recommend proceeding with caution, positioning it strictly as a baseline.
How to cite it
Written from CYP Inhibition Model's own documentation (model card). Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.