Literature · InstaDeepAI
ChatNT
Conversational model answering natural-language queries for DNA, RNA, and protein sequence tasks.
What ChatNT is
ChatNT is an 8-billion-parameter multimodal conversational model developed by InstaDeep. It allows researchers to interact with biological sequence data using natural-language instructions rather than dedicated command-line scripts or custom code.
The model addresses sequence annotation, classification, and numerical property estimation across DNA, RNA, and proteins. By translating biological sequence analysis into an interactive dialogue interface, it enables users to query genomic and molecular properties directly.
How it works
ChatNT uses a three-part architecture comprising an encoder, a cross-attention resampler, and a text decoder. The encoder is a 500-million-parameter Nucleotide Transformer v2 model pre-trained on genomes from 850 species, processing DNA broken into overlapping 6-mers up to 2,048 tokens (approximately 12 kilobases). A Perceiver Resampler then compresses the 2,048 sequence embeddings into 64 task-conditioned vectors using linear projection and gated attention.
These compressed sequence representations are inserted into user-defined placeholder locations within the text prompt, tokenised with an augmented LLaMA tokeniser. Finally, a frozen 7-billion-parameter Vicuna-7B decoder autoregressively generates natural-language answers, categorical labels, or numeric values. The system was instruction-tuned on 27 distinct biological tasks comprising 605 million DNA tokens (around 3.6 billion base pairs) and 273 million English tokens.
Key facts
- Total parameter count
- 8 billion
- Sequence encoder
- 500M Nucleotide Transformer v2
- Language decoder
- 7B Vicuna-7B (frozen)
- Sequence context length
- 2,048 tokens (approximately 12 kb)
- Training data size
- 605M DNA tokens, 273M English tokens across 27 tasks
- Required GPU memory
- 40 GB
What you can use it for
- Splice site detection
- Identifies biological features such as genomic acceptor splice sites from supplied sequence fragments via conversational prompts.
- Biological sequence classification
- Generates classification labels and multi-label descriptions for DNA, RNA, and protein sequences across diverse species and tissues.
- Numeric property prediction
- Estimates quantitative biological values conditioned on input sequences and natural-language queries.
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.
Runs on On-demand GPU
Runs on our servers with on-demand compute. A first run needs time to load the model; active capacity can be reused and scales down when idle.Run this model
ChatNT
Strengths and limits
Good at
- Processes long biological sequences up to roughly 12 kilobases using overlapping 6-mer tokenisation.
- Combines genomic representation learning with natural-language dialogue across DNA, RNA, and protein contexts.
- Enables non-computational researchers to query sequences without writing specialised bioinformatics scripts.
Watch out for
- Restricted to queries and formats related to the 27 biological tasks included in its instruction-tuning data.
- Not suitable for clinical or diagnostic workflows and requires experimental or bioinformatic verification.
- Prone to hallucinations on out-of-distribution inputs, and numerical predictions can exhibit digit-level errors.
- Requires at least 40 GB of GPU memory to run.
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 · See model card
The See model card licence carries terms worth reading before you depend on it.
Questions
Can I use ChatNT commercially?
Check the licence before commercial use. The See model card licence carries terms worth reading before you depend on it.
How do I run ChatNT 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 12s, and a run costs about 210 credits. Visitors without an account get 1,000 credits a month.
What sequence types and lengths can ChatNT process?
ChatNT processes DNA, RNA, and protein tasks, taking DNA sequences up to 2,048 overlapping 6-mer tokens, which equates to roughly 12 kilobases.
How are biological sequences supplied to the model?
Sequences are supplied alongside an English prompt containing dedicated placeholder tokens where the 64 resampled sequence embeddings are injected.
Can ChatNT generalize to any arbitrary biological sequence task?
No, the model is trained to answer questions on 27 specific biological tasks and may fail or hallucinate when applied outside of these domains.
Can ChatNT be used in a medical or clinical setting?
No, ChatNT is not a diagnostic or clinical tool, and critical outputs must be confirmed with wet-lab experiments or validated bioinformatics pipelines.
How to cite it
ChatNT: A Multimodal Conversational Agent for DNA, RNA and Protein Tasks, bioRxiv 2024.04.30.591835, 2024.
Written from ChatNT'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.