I develop interpretable machine learning frameworks that integrate multi-omic and spatial data to uncover the molecular logic of complex biological systems, with a focus on reproducibility, biological grounding, and transparent interpretation of high-dimensional genomics data.

My work focuses on developing interpretable and reproducible computational frameworks for genomics, uniting biological prior knowledge with multi-omic and spatial data. The central goal is to move beyond black-box prediction toward mechanistic understanding, building models that not only perform well but explain how genetic variation, regulatory programs, and perturbations reshape cellular states. Each framework emphasizes pathway- and network-level interpretability, cross-dataset generalization, and transparent benchmarking, establishing reusable analytical standards for single-cell and multimodal genomics. Through this approach, I aim to bridge machine learning, systems biology, and data-driven biological interpretation, supporting robust and generalizable discovery across biological contexts.

The following key projects are part of the MM-KPNN framework family, a unified effort to develop concept-bottleneck and biologically constrained models that embed prior knowledge directly into network architectures, ensuring interpretability, reproducibility, and mechanistic insight across multi-omic and spatial data.

A modular and interpretable graph framework for spatial transcriptomics in tissue microenvironments.

• Combines Graph Attention Networks (GAT) with knowledge-primed decoding
• Models cell–cell communication, immune exclusion, and stromal remodeling
• Outputs attention maps, pathway overlays, and ligand–receptor driver rankings

Interpretable multimodal neural network integrating scRNA-seq and scATAC-seq using biological priors.

• Decoder constrained by pathway and transcription factor nodes
• Enables mechanistic attribution of regulatory programs and cell states
• Designed for reproducible benchmarking across single-cell modalities

Pathway-bottleneck graph neural network for perturbation and drug-response prediction.

• Integrates multi-omic features with prior knowledge graphs
• Focuses on cross-dataset generalization across pharmacogenomic panels
• Provides pathway-level interpretability and reproducible evaluation

Concept-bottleneck framework for modeling drug and CRISPR perturbation responses at single-cell resolution.

• Implements pathway and TF bottlenecks for interpretability
• Measures attribution stability across perturbation conditions
• Supports counterfactual pathway analysis in single-cell datasets

A modular computational framework for the analysis of organoid systems.

• Addresses reproducibility, heterogeneity, and data integration
• Integrates RNA and protein modalities with interpretable ML
• Demonstrates end-to-end reproducibility through documented notebooks

Spatial mapping of tissue architecture using 10x Visium transcriptomics.

• Defines epithelial, immune, stromal, and proliferative regions
• Reveals spatial organization and regional heterogeneity
• Fully documented, end-to-end analytical workflow

End-to-end pipeline for structural variant discovery and annotation using PacBio long-read sequencing.

• Implements clinical annotation (ACMG/AMP) and variant filtering
• Includes functional scoring and visualization modules
• Designed for scalable deployment in HPC environments

Modular framework for rare-variant burden analysis in genomic cohorts.

• Supports SKAT, SKAT-O, and extended statistical methods
• Implements functional weighting and population correction
• Provides reproducible filtering and QC workflows

Systems biology workflow for reconstructing gene regulatory networks.

• Integrates TF–target priors with expression-based inference
• Performs network topology and modularity analysis
• Identifies functionally enriched regulatory modules

Gene co-expression analysis pipeline using WGCNA.

• Identifies expression modules and hub genes
• Evaluates biological function and module preservation
• Applies to bulk and single-cell RNA-seq datasets

Workflow for secure and efficient genomic data transfer using Globus.

• Supports HPC environments and structured data sharing
• Enables checksum validation and metadata tracking
• Designed for collaborative, reproducible research

Sally Yepes
📧 sallyepes233@gmail.com
🔗 GitHub: Sally332
🔗 Portfolio: sally332.github.io