Extended Data Fig. 2: PDGrapher efficiently predicts genetic perturbagens to shift cells from diseased to treated states in a random splitting setting within ten cell lines.
From: Combinatorial prediction of therapeutic perturbations using causally inspired neural networks
(a) PDGrapher provides accurate predictions for up to 1.09% more samples in the test set compared to the second-best baseline across Genetic-PPI datasets (genetic-PPI-ovary-ES2: 1.37% vs 0.28%). (b) scGen takes the leading position in nDCG across genetic-PPI datasets. (c) PDGrapher recovers ground-truth therapeutic targets at comparable rates as competing methods for genetic-PPI datasets. (d) PDGrapher has the best overall performance in perturbagen prediction within each cell line, evaluated by the averaged rank over multiple cell lines and metrics. The central line inside the box represents the median, while the top and bottom edges correspond to the first (Q1) and third (Q3) quartiles. The whiskers extend to the smallest and largest values within 1.5 times the interquartile range (IQR) from the quartiles. (e-g) Shown is the R2 of the response prediction module of PDGrapher compared to competing baselines for the top 20 (e), 40 (f), and 80 (g) differentially expressed (DE) genes.
