paolomanlapaz/rispice-base

RiSPICE Base

RiSPICE Base is the sequence classification model used by RiSPICE (Rice SNP Prioritization Integrating Chromatin Effects). It is built upon DNABERT-2-117M and provides the multi-label chromatin feature prediction head required for RiSPICE inference. Sequence-length-specific adaptation is provided separately through the RiSPICE LoRA adapters.

Model Summary

• Method: RiSPICE (Rice SNP Prioritization Integrating Chromatin Effects)
• Base Model: zhihan1996/DNABERT-2-117M
• Architecture: Sequence classification model
• Task: Multi-label chromatin feature prediction
• Number of Labels: 12

This repository is intended to be used together with one of the RiSPICE LoRA adapters:

• paolomanlapaz/rispice-1000bp
• paolomanlapaz/rispice-750bp
• paolomanlapaz/rispice-500bp

---

Feature Order

The model outputs logits (or probabilities after applying a sigmoid function) for the following 12 chromatin features, in the exact order shown below:

Index	Feature
0	ATAC-Seq
1	H3K4ac
2	H3K4me2
3	H3K4me3
4	H3K9ac
5	H3K9me1
6	H3K23ac
7	H3K27ac
8	H3K27me3
9	H3K36me3
10	H4K12ac
11	H4K16ac

---

Loading the Model

from transformers import AutoModelForSequenceClassification, AutoTokenizer

model = AutoModelForSequenceClassification.from_pretrained(
    "paolomanlapaz/rispice-base",
    trust_remote_code=True,
)

tokenizer = AutoTokenizer.from_pretrained(
    "paolomanlapaz/rispice-base",
    trust_remote_code=True,
)

After loading the base model, apply the desired RiSPICE LoRA adapter (500 bp, 750 bp, or 1000 bp) before performing inference.

---

Repository

For preprocessing, inference scripts, and documentation, please visit the RiSPICE repository:

GitHub: https://github.com/bioinfodlsu/rispice