DELIGHT-FM Epi-colon Age
Colon Tissue Age Prediction Model
1 min readupdated
This model solves a regression task for predicting tissue age from gene expression data in the colon. It predicts the log2(PCW) score (embryonic age in weeks) using single-cell RNA expression and enables verified perturbations (age-related changes) in genomic data. Based on the DELIGHT-FM family, the model uses a transformer architecture with Mamba extensions and was trained on tens of thousands of cells. This tool is useful for bioinformatics and genetics researchers analyzing cellular development.
Key features
- Retained regression head for predicting log2(PCW)
- Trained on a subset of over 85,000 cells
- Uses gene tokenization and specific resource files
- Includes notebooks for inference and validating age perturbations
What you can do
- Predict developmental age without retraining the model
- Run verified age perturbations according to the project notebooks
- Use the model alongside the base DELIGHT-FM Epi-colon repository for complex tasks
Example
from transformers import AutoModelForSequenceClassification
model = AutoModelForSequenceClassification.from_pretrained(
"chloehlee/delight-fm-epi-colon-age", trust_remote_code=True)
Read more
DetailsTakes gene token IDs and z-scored expression as input
Trained on 85,035 cells, validated on 25,155 cells
License: Apache 2.0
Requires Python 3.10, a PyTorch/CUDA 11.8 stack, and compiled Mamba extensions










Comments