CoolFace
Modelpublic

Sepideh2027/OmniLatent-TCGA-BRCA-model

sourceHugging Facemitupdated 8d agoView on Hugging Face
0likes
Model Card

Year: 2025–Present License: MIT Author: Sepideh Moafi

OmniLatent — Deep Representation Learning for TCGA-BRCA Transcriptomic Data

Model Description

OmniLatent is a deep autoencoder framework for learning compact, biologically meaningful latent representations from high-dimensional transcriptomic data.

The model was developed as part of the OmniLatent research project, which explores nonlinear representation learning for cancer genomics and precision medicine.

The model compresses genome-wide gene-expression profiles into a low-dimensional latent space, enabling downstream analysis such as clustering, visualization, and pattern discovery.

Model Details

  • —Model type: Stacked autoencoder
  • —Framework: PyTorch
  • —Latent dimension: 128
  • —Input dimension: 23,375 gene-expression features
  • —Dimensionality reduction: 99.5%
  • —Test reconstruction loss: 0.3129
  • —Format: PyTorch (best_model.pt)
  • —License: MIT

Intended Use

The model is intended for:

  • —Representation learning on bulk RNA-seq data
  • —Dimensionality reduction of transcriptomic profiles
  • —Clustering and visualization of cancer transcriptomes
  • —Research on cancer genomics and precision medicine
  • —Reproducible computational biology experiments

The model is not intended for clinical diagnosis, treatment decisions, or direct clinical use.

Training Data

  • —Dataset: TCGA-BRCA RNA-seq
  • —Samples: 1,231
  • —Genes: 23,375

Training

The model was trained as part of the OmniLatent research workflow. The associated project includes:

  • —End-to-end reproducible transcriptomic pipeline (data acquisition, preprocessing, gene filtering, normalization, QC)
  • —Stacked autoencoder architecture
  • —Latent space characterized via PCA, UMAP, and clustering
  • —Reproducible implementation and released research artifacts

Evaluation

Evaluation should be interpreted as research-model evaluation rather than clinical validation.

  • —Test reconstruction loss: 0.3129
  • —Latent space: 128-dimensional (99.5% reduction)

Performance depends on the downstream task and dataset. The model should be independently evaluated before use in downstream research applications.

Limitations

  • —Trained specifically on TCGA-BRCA data; may not generalize to other cancer types or datasets without retraining.
  • —Not validated for clinical decision-making.
  • —Latent representations should be interpreted with domain knowledge.
  • —Biological patterns identified in the latent space should be independently verified.

Ethical and Safety Considerations

This model is released for research and educational purposes. It should not be used as a substitute for professional medical or clinical expertise.

For applications involving human health data, outputs should be reviewed and independently verified by qualified researchers.

Reproducibility

Related resources:

  • —Model: Sepideh2027/OmniLatent-TCGA-BRCA-model
  • —Dataset: Sepideh2027/biolatent-brca-tcga

Citation

If you use this model in research, please cite the associated OmniLatent research output.

Contact

Sepideh Moafi GitHub: AIResearcher20 Hugging Face: Sepideh2027