Key takeaway
Drug response simulation requires a defined cell background, compound, dose, exposure time and endpoint. Expression changes, morphology and viability are distinct readouts and are not interchangeable.
Response is more than a score
Drug response can refer to expression changes, morphology, growth inhibition or other endpoints. Before modeling, define exactly which readout is being predicted to avoid conflating different outcomes.
Dose sets the comparison context
The same compound can produce different response patterns at different concentrations. Retain dose units, vehicle controls, batches and the tested concentration range. Doses outside the observed range are extrapolation scenarios that need further validation.
Timing shapes what is observed
Early signals and later state changes are not equivalent. A single endpoint cannot easily distinguish transient stress, sustained responses and adaptation. Studies of dynamic processes need appropriate observation time points.
Cellular background matters
Drug responses depend on cell type, genetic background and culture environment. Report performance for novel contexts separately from contexts observed during training so aggregate metrics do not conceal subgroup failures.
Translate predictions into experiments
Prioritize conditions with clear predicted differences, sufficient evidence and relevance to the study, while retaining positive and negative controls. Mechanistic interpretations need independent phenotypes or orthogonal experiments rather than expression correlation alone.
Further reading
Arc Institute · Virtual Cell Initiative ↗
Tahoe · Data and research approach ↗
View all references
Sources and evidence
Perturbation data, models and research tools connect biological questions with experimental evidence. Tahoe · Research & data ↗
Evaluation across datasets, perturbations and cellular contexts supports task-specific model selection. Nature Methods · Generalizable perturbation prediction ↗