Predictive Models

The Predict section of the BioLM Console provides an interactive catalog of available predictive models for protein and nucleic acid analysis. These models can be used for sequence folding, mutation prediction, annotation, and more.

Each model tile displays:

Model Name – the identifier used throughout the platform and API.
Description – a short summary of the model’s intended use case.
View API Docs – a direct link to the model’s API documentation, where you can learn how to call the model programmatically using the BioLM SDK or REST API.
Cold Start Toggle – an indicator showing whether a model instance is currently active or will need to be spun up before use. Cold starts may add slight delay when running predictions.

Browsing Models

Models are organized in a grid view for quick discovery. You can scroll through the available models or use the Tags menu in the upper right to filter by category or function (for example, “antibody design,” “protein folding,” or “DNA analysis”).

Example Model Types

Some of the commonly used models include:

AbLang-2 – predicts antibody mutations to enhance specificity and affinity.
ABodyBuilder3 pLDDT – predicts 3D antibody structures and per-residue confidence scores.
AlphaFold2 – predicts general protein 3D structures from amino acid sequences.
DNABERT-2 – generates embeddings and log probabilities for DNA sequences.
BioLMTox-2 – screens peptides and proteins for toxin potential.

Using a Model

Select a model from the catalog.
Click View API Docs to open detailed usage instructions, including:
- Input format (e.g., FASTA sequence, SMILES, etc.)
- Output schema and example results
- Code snippets for Python SDK and REST API
Optionally toggle Cold Start if you wish to manually pre-start a model before calling it through the API.

Typical Use Cases

Folding or structure prediction of new protein sequences
Predicting mutation likelihoods or structural confidence scores
Screening for toxicity or off-target risk
Embedding biological sequences for downstream machine learning tasks

Notes

Predictive models are accessible through both the web console and the BioLM SDK.
Some models may have usage limits depending on your subscription plan or available budget balance (visible in the console header).