seqme 0.2.3

Metrics for evaluating biological sequence design.

seqme is a modular and extendable python library containing model-agnostic metrics for evaluating biological sequence designs. It enables benchmarking and comparison of generative models for small molecules, DNA, RNA, peptides, and proteins.

Key features:

• Metrics: A collection of sequence-, embedding-, and property-based metrics for evaluating generative models designs.
• Models: Out-of-the-box, pre-trained property and embedding models for small molecules, DNA, RNA, peptides, and proteins.
• Visualizations: Functionality to display metric results from single-shot and iterative optimization methods as tables and plots.

Is a metric or model missing? seqme's modular metric and third-party model interfaces make adding your own easy.

Installation

You need to have Python 3.10 or newer installed on your system. To install the base package do:

$ pip install seqme

To install sequence-specific models as well, include the appropriate extras specifiers. Check the individual model docs for installation instructions.

Quick start

Install seqme and the protein language model, ESM-2.

$ pip install 'seqme[esm2]'

Run in a Jupyter notebook:

import seqme as sm

sequences = {
    "Random": ["MKQW", "RKSPL"],
    "UniProt": ["KKWQ", "RKSPL", "RASD"],
    "HydrAMP": ["MMRK", "RKSPL", "RRLSK", "RRLSK"],
}

cache = sm.Cache(
    models={"esm2": sm.models.ESM2(
        model_name="facebook/esm2_t6_8M_UR50D", batch_size=256, device="cpu")
    }
)

metrics = [
    sm.metrics.Uniqueness(),
    sm.metrics.Novelty(reference=sequences["UniProt"]),
    sm.metrics.FBD(reference=sequences["Random"], embedder=cache.model("esm2")),
]

df = sm.evaluate(sequences, metrics)
sm.show(df) # Note: Will only display the table in a notebook.

Check out the docs for in-depth tutorials and examples.

Citation

Preprint is coming soon.