Computational Biologist | Cancer Genomics & Interpretable AI
I develop interpretable machine learning frameworks that integrate multi-omic and spatial data to uncover the molecular logic of cancer. My work combines mechanistic interpretability, reproducibility, and biological grounding, aiming to build research-ready frameworks rather than one-off analyses. Current focus areas include spatial transcriptomics, drug-response modeling, and graph-based multimodal architectures.
SpatialMMKPNN
Interpretable graph neural network for spatial transcriptomics
Combines Graph Attention Networks and knowledge-primed decoding to reveal spatial signaling in the tumor microenvironment.
DOI: 10.5281/zenodo.17189130
MM-KPNN
Multimodal neural network constrained by biological priors
Integrates scRNA-seq and scATAC-seq data through pathway and TF bottlenecks for interpretable cell-state modeling.
DOI: 10.5281/zenodo.17194732
DrugResponse-GNN
Graph neural network for drug sensitivity prediction
Uses pathway bottlenecks to model cross-panel pharmacogenomic generalization across CCLE, GDSC, and NCI-60 datasets.
DOI: 10.5281/zenodo.17189237
Perturbation-MMKPNN
Concept-bottleneck model for single-cell perturbation data
Interpretable modeling of transcriptional responses to drug and CRISPR perturbations across multiple datasets.
DOI: 10.5281/zenodo.17189224
Spatial Transcriptomic Mapping
Biologically grounded mapping of tumor architecture through 10x Visium data
Region-based analysis revealing tumor, stromal, and immune organization across primary and metastatic samples.
DOI: 10.5281/zenodo.17272950
Interpretable by design · Reproducible by construction · Generalizable across data modalities.
My long-term goal is to establish transparent and reusable computational standards for cancer genomics frameworks that predict, explain, and generalize across tissues, datasets, and modalities.
📧 sallyepes233@gmail.com
🔗 GitHub Profile