The concept of the "virtual cell" aims to create a comprehensive model that maps healthy cells and diseases, allowing researchers to visualize treatment paths back to a healthy state. The current focus is on the transcriptome, utilizing large public datasets that include cell lines with genetic and drug perturbations. There’s a push to develop a foundational model that can integrate this data, similar to how AlphaFold transformed protein structure prediction, but the author believes the journey will resemble the evolution of Genome Wide Association Studies (GWAS) rather than a direct replication of AlphaFold's success. Excitement in this field stems from major advancements in data collection methods, particularly in RNA sequencing and imaging. However, significant challenges remain. Experimental batch effects can introduce non-biological variability, complicating data interpretation. Moreover, current metrics for assessing model quality are inadequate, often focusing on dataset reconstruction rather than genuine biological insights. The author draws parallels to GWAS, where researchers have learned to navigate noisy data and isolate confounding factors to establish clearer correlations between genetic variants and phenotypes, aided by large datasets like the UK Biobank. The author suggests that successful application of transcriptome models will require operators to extract useful insights from public datasets while tailoring their research to specific biological questions. This means constructing focused datasets that can address particular diseases or cell lines, rather than trying to create one-size-fits-all models. Emphasizing the importance of isolating confounding noise and developing better evaluation metrics will be essential. Ultimately, the ability to distill large datasets into targeted biological questions will drive progress in this area, rather than simply accumulating vast amounts of data.