Drug target prediction through deep learning functional representation of gene signatures
Problem
Machine learning analyses of gene signatures rely on matching gene identities and ignore preexisting knowledge of gene functions, limiting compound-target prediction and mechanism-of-action studies.
Why it matters
Better compound-target prediction supports target deconvolution and mechanism-of-action studies in drug development.
What the AI did
FRoGS trains a deep learning model to represent gene signatures projected onto biological functions (analogous to word2vec), applied to the Broad Institute's L1000 datasets.
Results
More effective compound-target predictions than gene-identity-based models; significantly more high-quality compound-target predictions than existing approaches (no numbers given).
Impact
Improves compound-target prediction effectiveness over existing gene-identity approaches, per the abstract.
Record
| Venue | Nature Communications |
|---|---|
| Published | 2024 |
| Work type | article |
| Authors | Hao Chen, Frederick J. King, Bin Zhou, Yu Wang, Carter J. Canedy, Joel Hayashi, Yang Zhong, Max W. Chang et al. |
| Citations | 95 (OpenAlex, at import time) |
| References | 83 |
| Model or method | FRoGS (deep learning functional representation model) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41467-024-46089-y |
| Open access | gold |
| Keywords | drug target prediction, gene signatures, deep learning, word2vec, L1000, functional representation |
Importance rating
real but narrower advance
Improves compound-target prediction and mechanism-of-action studies in drug development bioinformatics; abstract claims general utility but gives no scale or impact numbers.