< Back to Article

Connectivity Homology Enables Inter-Species Network Models of Synthetic Lethality

Fig 2

A. Receiver operating characteristic (ROC) curves for classification of SL/non-SL gene in S. pombe using S. cerevisiae as source.

Comparison of untranslated (“raw”) parameters (gray, AUC = 0.67) and the translated parameters used in SINaTRA (red, AUC = 0.86). B. ROC curve of SL predictions using SINaTRA (AUC = 0.86) compared functional homology of gene pair products (AUC = 0.81) and gene homology (AUC = 0.60). The model based on gene homology was created using only gene pairs with homology data. C. Positive predictive value (PPV) of all (dark gray) and within-complex (red) gene pairs. When accounting for the expected ratio of SL:NSL (1:1000), a SINaTRA score threshold of 0.95 yields a median PPV of 17% (a 170-fold increase over what is expected by chance). At 0.85, the PPV drops to 7%. PPV increases in within-complex gene pairs, suggesting that this may be a good initial filter for experimental validation. D. At each SINaTRA score cutoff, we plot the number of experimentally identified SL pairs in that bin (red), as well as the number we expect to find at each level (gray). E. SINaTRA scores of all human predictions, as well as pairs predicted or found to be SL in two datasets: DAISY and Syn-Lethality. F. We compare the predictive ability of SINaTRA score to identify genes belonging to DAISY (tested) and Syn-Lethality datasets compared to functional similarity and homology.

Fig 2