FishNALM-20L_H3K27me3

FishNALM-20L_H3K27me3 is a fine-tuned version of FishNALM-20L_pretrain for H3K27me3 prediction in fish genomics.

Model description

This repository contains a task-specific fine-tuned checkpoint from the FishNALM model family. The model was initialized from the pretrained base model FishNALM-20L_pretrain and then fine-tuned for H3K27me3 prediction.

Task

Task name: H3K27me3 prediction
Task type: binary classification
Prediction target: H3K27me3-positive vs H3K27me3-negative genomic sequences

Examples:

• CTCF TFBS prediction
• Pou5f1 TFBS prediction
• Sox2 TFBS prediction
• histone modification prediction
• promoter prediction
• splice donor prediction
• splice acceptor prediction
• splice classification

Base model

• Base model repository: xia-lab/FishNALM-20L_pretrain
• Model family: FishNALM
• Initialization type: pretrained checkpoint + downstream fine-tuning

Training data

This model was fine-tuned on H3K27me3 prediction data from FishGUE.

Evaluation

• Primary metric: MCC
• Evaluation split / strategy: predefined train/validation/test split

Intended uses

This model is intended for:

• fish genomics sequence classification
• downstream task inference on sequences similar to the fine-tuning setting
• comparative benchmarking within fish genomic prediction tasks

Limitations

• This is a task-specific fine-tuned model and should be used within the scope of H3K27me3 prediction.
• Generalization to other species, tasks, or sequence lengths may be limited.
• This is a research model and is not intended for clinical or diagnostic use.

How to use

Load tokenizer and model

from transformers import AutoTokenizer, AutoModelForSequenceClassification

repo_name = "xia-lab/FishNALM-20L_H3K27me3"

tokenizer = AutoTokenizer.from_pretrained(repo_name)
model = AutoModelForSequenceClassification.from_pretrained(repo_name)

Example inference

import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification

repo_name = "xia-lab/FishNALM-20L_H3K27me3"
sequence = "ATGCGTACGTTAGCTAGCTAGCTAGCTAGCTA"

tokenizer = AutoTokenizer.from_pretrained(repo_name)
model = AutoModelForSequenceClassification.from_pretrained(repo_name)

inputs = tokenizer(
    sequence,
    return_tensors="pt",
    truncation=True,
    padding="max_length",
    max_length=512,
)

with torch.no_grad():
    outputs = model(**inputs)

logits = outputs.logits
probabilities = torch.softmax(logits, dim=-1)
prediction = torch.argmax(probabilities, dim=-1)

print("logits:", logits)
print("probabilities:", probabilities)
print("prediction:", prediction)

Label mapping

• 0: negative
• 1: positive

Files in this repository

Typical files in this repository may include:

• config.json
• model.safetensors
• tokenizer.json
• tokenizer_config.json
• special_tokens_map.json
• vocab.txt
• README.md

Citation

If you use this model, please cite the FishNALM manuscript.

Contact

For questions, please contact: [email protected]