Working with SeqFrame

Sequence-centric dataframes over Parquet, with DuckDB-backed query and enrichment.

Install the optional extra:

bash

$ pip install "biolm-sdk[seqframe]"

(The same DuckDB / pandas / pyarrow stack is also included in biolm-sdk[pipeline].)

A SeqFrame is to biological sequences what a DataFrame is to tabular data. It is complementary to Managing datasets: datasets inventory bags of files; SeqFrame queries, filters, and enriches sequence tables.

Quick start

python

from biolm import SeqFrame

sf = SeqFrame.from_fasta("proteins.fasta")
sf = sf.query.filter("length < 300").query.limit(10)
sf.io.to_parquet("filtered.parquet")
sf2 = SeqFrame.read("filtered.parquet")

Namespaced APIs

sf.query — filter, select, join, sort, limit, group_by
sf.io — to_fasta, to_csv, to_jsonl, to_parquet
sf.bio — length / type helpers (translate needs biopython)
sf.models — predict / embed via the BioLM API
sf.protocols — run a protocol and join results (default join key id; query.join defaults to sequence_hash)
sf.lab — LLTP convert helpers (to_lltp / from_lltp / merge); orchestration is Lab-in-the-Loop (LLTP) (biolm.lab / biolm lab)

Datasets bridge

Write a SeqFrame into a local dataset (type: seqframe) and open it again:

python

from biolm.datasets import DatasetClient

client = DatasetClient()
ds = sf.to_dataset("my-proteins", client=client, tags=["design"])
sf2 = ds.open_seqframe()
# or: SeqFrame.from_dataset(ds)

Resolution uses attrs.seqframe_path when set, otherwise exactly one .parquet file under the dataset.

Parquet metadata (seqframe.version / seqframe.schema) is documented in SeqFrame Metadata Schema Reference.

Lab-in-the-loop (LLTP) submit / status / results live in Lab-in-the-Loop (LLTP).

See biolm.seqframe package and biolm.seqframe.