Cell morphological representations of genes enhance prediction of drug targets
Iyer, N. S.; Michael, D. J.; Chi, S.-Y. G.; Arevalo, J.; Chandrasekaran, S. N.; Carpenter, A. E.; Rajpurkar, P.; Singh, S.
Show abstract
Identifying a chemicals mechanism of action on biological systems is critical for drug development. A compounds mechanism can sometimes be identified by matching its image-based morphological profile to a well-annotated library. We enhance this approach by incorporating gene representations, demonstrating improved classification performance compared to direct matching methods. These representations are generated using morphological profiles of cells with artificially altered gene expression. A transformer model is trained to classify gene-compound pairs as true or false interactions, using both gene and compound profiles. The model then ranks likely target genes for unseen compounds. The strategy performs well for compounds targeting known genes but has limited effectiveness for novel targets, likely due to the current dataset size. However, the performance increase over existing profile-matching methods is notable. Future work with expanded datasets may enhance predictive capabilities, potentially accelerating early-stage drug discovery by improving target identification for novel compounds.
Matching journals
The top 9 journals account for 50% of the predicted probability mass.
Similar papers in this journal
- Capturing cell heterogeneity in representations of cell populations for image-based profiling using contrastive learning 96%
- Predicting drug polypharmacology from cell morphology readouts using variational autoencoder latent space arithmetic 95%
- A machine learning and network framework to discover new indications for small molecules 95%
Similar papers in this journal
- Compound activity prediction with dose-dependent transcriptomic profiles and deep learning 97%
- Adding stochastic negative examples into machine learning improves molecular bioactivity prediction 96%
- DrugHIVE: Target-specific spatial drug design and optimization with a hierarchical generative model 95%
Similar papers in this journal
- Merging Bioactivity Predictions from Cell Morphology and Chemical Fingerprint Models Using Similarity to Training Data 96%
- DrugDiff - small molecule diffusion model with flexible guidance towards molecular properties 95%
- Stereochemically-aware bioactivity descriptors for uncharacterized chemical compounds 94%
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.