WAVE predicts a cell’s gene-expression response from its starting state and a compound

The Nature Communications accepted article appeared on September 18. The aim is to prioritize experiments across many compounds and cellular settings; a predicted transcriptional response is not a prediction of clinical benefit. [1]

The public code makes the transformation understandable. One component compresses the starting expression profile, another represents the compound and a third predicts a change that is added to the reconstructed baseline. The repository provides data-format, training and evaluation instructions, including a measure of how well the predicted change matches the observed change. [2]

The paper evaluates bulk and single-cell datasets and includes laboratory follow-up in lung-cancer cell lines. Those experiments provide evidence beyond a computational score. Important limits remain: WAVE does not explicitly model dose-response or time-series dynamics, and the authors call for stronger validation on unfamiliar compounds and cellular contexts. A laboratory dose-response experiment in the paper does not mean the prediction model itself has acquired a dose axis. [3]

Reserve an entire compound set or cellular context from development, predict its responses in advance and compare with measurements. Judge the ranking against a simple baseline and the laboratory’s existing selection process. That tests whether WAVE helps choose a better next experiment.