The generated representations will be evaluated in a compound classification task using the kNN algorithm in 5-fold cross-validation. Metrics will be calculated using inter and intra cell line matching. Train/test split will be undisclosed but follows plate-wise split. Evaluation will be performed using only compound plates of 2020_11_04_CPJUMP1 dataset and will omit DMSO well but any data can be used to develop and train a method. Participants will also be asked to send a paper describing the methodology (up to 4 pages without citations).

Inter and intra cell line matching:

- Train using U2OS data, evaluate using U2OS data.
- Train using U2OS data, evaluate using A549 data.
- Train using A549 data, evaluate using U2OS data.
- Train using A549 data, evaluate using A549 data.