Literature · google
TxGemma 2B
A 2-billion-parameter language model fine-tuned for therapeutic property prediction across drug discovery tasks.
What TxGemma 2B is
TxGemma 2B is a lightweight, open-weight decoder-only language model developed by Google. Fine-tuned from Gemma 2, it is tailored specifically for computational therapeutic development and drug discovery workflows.
The model targets diverse biological entities and modalities, including small molecules, proteins, nucleic acids, diseases, and cell lines. It predicts therapeutic properties by processing biological and chemical representations such as SMILES strings, amino acid sequences, and nucleotide sequences directly in structured text prompts.
Trained on curated datasets from the Therapeutics Data Commons, TxGemma 2B is designed to provide competitive predictive capabilities at a compact size, offering a practical foundation for specialized property prediction and downstream fine-tuning.
How it works
TxGemma 2B uses a decoder-only transformer architecture built directly upon Gemma 2. It was trained using JAX to leverage modern TPU hardware infrastructure.
The model underwent instruction fine-tuning using a collection of therapeutic datasets (TxT) sourced from the Therapeutics Data Commons (TDC). Unlike conversational variants, this prediction-oriented variant is optimised for structured prediction queries using specific TDC-style prompt templates rather than open multi-turn dialogue.
Key facts
- Base architecture
- Gemma 2 (decoder-only transformer)
- Parameter count
- 2 billion
- Model version
- 1.0.0
- Creation date
- 18 March 2025
- Training software
- JAX
- Input modalities
- Text (including SMILES, amino acid sequences, nucleotide sequences)
- Output modality
- Text
What you can use it for
- Blood–brain barrier permeability prediction
- Predict whether a small molecule crosses the blood–brain barrier based on its SMILES string using standardized TDC prompts.
- Drug–target interaction screening
- Evaluate potential interactions between candidate therapeutic compounds and specific protein or nucleic acid targets.
- Clinical trial approval forecasting
- Predict outcomes related to therapeutic advancement and clinical trial approval for potential drug candidates.
- Custom therapeutic fine-tuning
- Serve as a pre-trained base model for further fine-tuning on proprietary internal assay measurements and drug discovery datasets.
Run it here
We list this model but cannot run it yet: excluded (gated): HAI-DEF terms, as above..
Everything on this page still applies. Running it needs a loader for this repository, not a bigger machine.
Strengths and limits
Good at
- Achieves competitive performance on therapeutic tasks with a compact 2B parameter footprint.
- Processes diverse biological representations, including SMILES, nucleotide, and amino acid sequences.
- Acts as an efficient foundation for further domain-specific fine-tuning on internal laboratory data.
- Can be integrated into larger agentic workflows powered by host reasoning models.
Watch out for
- Requires strictly formatted TDC prompts consisting of instructions, context, question, and candidate inputs rather than free-form conversational interaction.
- Does not include the conversational or explanatory capabilities present in larger TxGemma-Chat versions.
- Trained only on public data from the Therapeutics Data Commons with commercial-friendly licensing, unlike experimental models described in the underlying research manuscript.
- Requires independent downstream validation on representative domain data prior to application in specific operational settings.
Inputs and outputs
You provide
Prompt
The model continues from here.
You get back
The model's completion.
Licence
Check the licence before commercial use · other
This model is gated by its authors — its licence has to be accepted on the source repository before it will run.
Questions
Can I use TxGemma 2B commercially?
Check the licence before commercial use. This model is gated by its authors — its licence has to be accepted on the source repository before it will run.
How does TxGemma 2B differ from the TxGemma 9B and 27B chat models?
TxGemma 2B is a dedicated prediction model tuned for direct task outputs using narrow prompting structures. It lacks the multi-turn conversational abilities and reasoning explanations provided by the larger chat models.
What input formats are supported by TxGemma 2B?
The model accepts textual input formatted according to TDC task templates, which can incorporate small-molecule SMILES strings, protein amino acid sequences, nucleotide sequences, and natural language descriptions.
What data was used to train this public version of TxGemma 2B?
The released model was trained on instruction-tuning datasets from the Therapeutics Data Commons (TDC) that carry commercial licenses, spanning over 15 million data points across various therapeutic modalities.
How to cite it
Wang, Eric and Schmidgall, Samuel and Jaeger, Paul F. and Zhang, Fan and Pilgrim, Rory and Matias, Yossi and Barral, Joelle and Fleet, David and Azizi, Shekoofeh. TxGemma: Efficient and Agentic LLMs for Therapeutics, 2025. arXiv:2504.06196.
Written from TxGemma 2B'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.