Multi-Modal Foundation · IBM Research
BioVERSE
A multimodal model answering natural-language questions about proteins and small molecules from their embeddings.
What BioVERSE is
BioVERSE is a multimodal system developed by IBM Research designed for cross-modal question answering. It bridges biological and chemical representations directly into a large language model, allowing researchers to query biological entities using natural language.
Standard language models frequently rely on internal textual memory that may hallucinate or miss obscure molecular details. BioVERSE addresses this by grounding natural-language answers directly in biological entity embeddings rather than depending solely on text learned during training.
The system takes either an amino acid sequence or a chemical SMILES string alongside a text question, producing factual, grounded textual answers regarding the specified protein or small molecule.
How it works
BioVERSE bridges MAMMAL bio-entity embeddings into the representation space of an IBM Granite large language model. By mapping molecular and sequence representations into the language model's feature space, it enables zero-shot cross-modal question answering.
The model receives an entity input—such as a protein sequence or small-molecule SMILES string—and a user question in plain text. It projects the biological embedding so the language model can generate responses directly informed by the underlying entity representation.
Key facts
- Version
- 1.0
- Supported modalities
- Protein sequence, small molecule, text
- Primary task
- Cross-modal question answering
- Underlying models
- MAMMAL embeddings bridged into Granite LLM
- GPU memory required
- 40 GB
What you can use it for
- Protein sequence querying
- Analyse protein sequences by providing an amino acid sequence and querying specific functional or structural characteristics in natural language.
- Small-molecule interrogation
- Submit small molecules using SMILES notation to ask grounded biological and chemical questions.
- Zero-shot biological Q&A
- Obtain answers to novel biological questions about entities without requiring task-specific fine-tuning on those entities.
Run it here
No installation and no GPU of your own: fill in the inputs and it runs on our servers. Results stay in your browser unless you sign in.
Not runnable yet
This model isn't available to run right now.Run this model
BioVERSE
Strengths and limits
Good at
- Grounds responses in bio-entity embeddings rather than relying solely on the language model's training memory.
- Enables zero-shot cross-modal question answering across both small molecules and protein sequences.
- Integrates specialised MAMMAL molecular representations into a Granite language model space.
Watch out for
- Requires substantial hardware resources, needing 40 GB of GPU memory.
- Inputs are restricted to single entity sequences or SMILES strings alongside a text prompt.
- Modality support is limited strictly to protein sequences, small molecules, and text.
Inputs and outputs
You provide
Entity (sequence or SMILES)
Question
You get back
Grounded response.
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 BioVERSE commercially?
Check the licence before commercial use. The Apache-2.0 licence carries terms worth reading before you depend on it.
How do I run BioVERSE without installing anything?
Use the form under “Run it here” on this page: fill in the inputs and press Run model. It runs on our servers on 1× A100 40GB, typically in 9.0s, and a run costs about 84 credits. Visitors without an account get 1,000 credits a month.
What input formats does BioVERSE accept for biological entities?
It accepts biological entities entered as protein amino acid sequences or as small-molecule SMILES strings.
Which underlying language and embedding models are used?
BioVERSE uses MAMMAL embeddings bridged into a Granite large language model.
What is the computational requirement to run BioVERSE?
The model requires 40 GB of GPU memory.
Written from BioVERSE's own documentation. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.