Fig. 3: Retrospective study design and results.
From: A Bayesian active learning platform for scalable combination drug screens
a Retrospective studies are conducted by processing an existing dataset into candidate row/column plates of the kind considered in this work and simulating the choices made by both the Random and BATCHIE data collection procedures. After data collection, the models are trained on the data collected by each of the procedures and evaluated for accuracy. Created in BioRender. Tansey, W. (2024) BioRender.com/i97p897. b Statistics for the datasets used in our retrospective studies. Created in BioRender. Tansey, W. (2024) BioRender.com/e57c809. c Heatmaps showing the fraction of possible experiments observed for each drug combination. d BATCHIE outperforms Random when both are evaluated after 15 rounds of data collection; p-values derived from a two-sided Mann-Whitney U-Test with no adjustments made for multiple comparisons. e The number of additional experiments needed for Random to achieve comparable performance to BATCHIE grows with the number of BATCHIE rounds across all datasets. Lines represent mean values; error bars denote standard deviations. f The number of additional batches needed for Random to achieve comparable performance to BATCHIE grows with the size of the cell line library; ρspear is Spearman’s rho for nonparametric rank correlation with a two-sided test for the p-value with no adjustments made for multiple comparisons. For boxplots in (d, f), center lines denote means, box limits denote standard deviations, and whiskers denote extremal values. d–f n = 25 replicates for all plots. Source data are provided as a Source Data file.
