Georgia Institute of Technology
Structure-guided Genetics-informed Embeddings Enable Zero-shot Drug Sensitivity Prediction
Abstract
dc:description.abstractAccurate prediction of anticancer drug response remains a central challenge in precision oncology, as experimental screening across diverse cell lines is costly and time-intensive1,2. Large-scale pharmacogenomic resources such as the Genomics of Drug Sensitivity in Cancer3 enable systematic analysis of drug–cell line interactions; however, most computational approaches require response data for each drug, limiting generalization to novel compounds4. Here, we propose a graph neural network framework that learns drug representations directly from molecular structure, supervised by pharmacogenomic similarity derived from response profiles. By aligning structural embeddings with biological response similarity, the model captures meaningful relationships between drugs while remaining independent of cell line features at inference. This enables zero-shot (cold-start) prediction, allowing similarity estimation and downstream prioritization for previously unseen drugs using only their molecular graphs. Our model was evaluated on the data from the Genomics of Drug Sensitivity in Cancer (GDSC)3 datasets accessed via PyTDC v1.1.155. It achieves substantial improvements in functional similarity prediction (Pearson=0.423, NDCG@10=0.480) and enables structure-only zero-shot twin matching of sensitive cell lines (average 47% Precision@15, up to 100% in top cases), demonstrating its utility for guiding preclinical experimental design in precision oncology.
Degree
thesis:*- Name thesis:degree_name
- Biomedical Engineering, MS
- Department thesis:degree_department
- Biomedical Engineering (Joint GT/Emory Department)
- Grantor
- Georgia Institute of Technology
- Year dc:date.issued
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Gehlot, Pranjal
- Advisor dc:contributor.advisor
-
- Pai, Balakrishna
- Committee members dc:contributor.committeemember
-
- Wang, May
- Coskun, Ahmet
- ORCID person:identifier.orcid
Rights
- Language dc:language.iso
- English
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/1853/81602