Predicting Cellular Responses to Novel Drug Perturbations at a Single-Cell Resolution
Authors: Leon Hetzel, Simon Boehm, Niki Kilbertus, Stephan Günnemann, mohammad lotfollahi, Fabian Theis
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We introduce chem CPA, a new encoder-decoder architecture to study the perturbational effects of unseen drugs. We combine the model with an architecture surgery for transfer learning and demonstrate how training on existing bulk RNA HTS datasets can improve generalisation performance. Better generalisation reduces the need for extensive and costly screens at single-cell resolution. We envision that our proposed method will facilitate more efficient experiment designs through its ability to generate in-silico hypotheses, ultimately accelerating drug discovery. Our main contributions are: ... 3. We show how chem CPA outperforms existing methods on the task of predicting unobserved drug-covariate combinations. At the same time, we demonstrate chem CPA s versatility and evaluate chem CPA on generalisation tasks that cannot be modeled using any previously existing method. |
| Researcher Affiliation | Academia | Leon Hetzel 1, 3, Simon Böhm 3, Niki Kilbertus2, 4, Stephan Günnemann2, Mohammad Lotfollahi1, 5, and Fabian Theis1, 3 {leon.hetzel, simon.boehm, niki.kilbertus}@helmholtz-muenchen.de s.guennemann@tum.de, {mohammad.lotfollahi, fabian.theis}@helmholtz-muenchen.de 1Department of Mathematics, Technical University of Munich 2Department of Computer Science, Technical University of Munich 3Helmholtz Center for Computational Health, Munich 4Helmholtz AI, Munich 5 Wellcome Sanger Institute, Cambridge |
| Pseudocode | No | The paper provides architectural diagrams (Figure 1) and mathematical formulations of the model, but it does not include any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code is available at github.com/theislab/chem CPA. |
| Open Datasets | Yes | We use the sci-Plex3 (Srivatsan et al., 2020) and the L1000 (Subramanian et al., 2017) datasets for the main evaluation on single-cell data and pretraining on bulk experiments, respectively. |
| Dataset Splits | No | The paper mentions creating |
| Hardware Specification | No | The paper does not provide any specific hardware details such as GPU models, CPU types, or memory specifications used for running the experiments. It only discusses the experimental setup at a general level. |
| Software Dependencies | No | The paper mentions using "RDKit features" for the molecule encoder but does not specify its version. It does not list any other software libraries, frameworks, or their specific version numbers that would be necessary for reproducibility. |
| Experiment Setup | No | The paper states, |