"Multimodal Graph Representation Learning for Single-Cell Cancer Genomics"
Xian F. Mallory
Department of Computer Science,
Florida State University (FSU)
Wednesday, Oct 7, 2026 
- Colloquium - 499 DSL Seminar Room
- 03:30 to 04:30 PM Eastern Time (US and Canada)
Abstract:
Single-cell sequencing makes it possible to characterize tumor heterogeneity at cellular resolution. However, each molecular modality has inherent limitations. Transcriptomic profiles of malignant and normal cells often overlap, copy number alteration (CNA) signals are low resolution and not comprehensive, and single-nucleotide variations (SNVs) are largely unobserved in low-coverage sequencing. This talk presents a graph-based framework for integrating complementary modalities. Cells are represented as nodes, one modality defines the node features, and a second modality defines the edge weights. A graph attention autoencoder then learns low-dimensional cell embeddings that jointly reflect both sources of information. Two applications of this framework will be discussed. SCGclust performs subclone clustering on ultra-shallow single-cell DNA sequencing data by combining read-count-based CNA features with SNV-derived cell similarity. It consistently outperforms methods that rely on a single signal. DeepMalignant identifies malignant cells in single-cell RNA sequencing data by combining expression signatures with CNA-derived cell similarity. Across 26 samples, four cancer types, and three sequencing platforms, it achieves the most favorable balance of precision and recall, and it generalizes to spatial transcriptomics. Ablation analyses confirm that both the multimodal edge construction and the attention mechanism contribute to performance.
Speaker Biography:
Dr. Mallory is interested in developing novel computational techniques for analyzing large data sets in order to understand biological phenomena and processes, with particular emphasis on cancer. Dr. Mallory’s work draws upon methodologies from combinatorial optimization, statistical inference, and machine learning.
| File | Description | File size |
|---|---|---|
 mallory.png | Headshot | 676 kB |
