Skip to content

DNALLM Documentation

Prediction Data

zhangtaolab/DNALLM

Prediction Data Preparation¶

This tutorial explains how to prepare input data for inference tasks with DNALLM.

Full Notebook¶

View Full Notebook

Prerequisites¶

uv pip install -e '.[base,inference,cuda124]'

Supported Input Formats¶

DNALLM inference accepts:

Plain text files (.txt, .fasta) — one sequence per line
CSV files — with sequence and optional label columns
JSON files — with structured data fields

Data Requirements¶

Sequences should use standard nucleotide characters: A, T, C, G
Remove or replace ambiguous characters (N, R, Y, etc.) before inference
Ensure consistent sequence lengths if required by your model
Use UTF-8 encoding for all text files

Example: Inference from CSV¶

from dnallm import load_config, load_model_and_tokenizer
from dnallm.inference import DNAInference

configs = load_config("inference_config.yaml")
model, tokenizer = load_model_and_tokenizer(
    model_name="zhangtaolab/plant-dnabert-BPE",
    task_config=configs['task'],
    source="huggingface"
)

inferencer = DNAInference(
    model=model,
    tokenizer=tokenizer,
    config=configs
)

results = inferencer.infer(
    file_path="../../../../tests/test_data/binary_classification/test.csv",
    batch_size=32
)