Overview
Identification of the most appropriate pharmacotherapy for each
patient based on genomic alterations is a major challenge in
personalized oncology. PANACEA is a collection of
personalized anti-cancer drug prioritization approaches utilizing
network methods. The methods utilize personalized “driverness” scores
from driveR to
rank drugs, mapping these onto a protein-protein interaction network
(PIN). The “distance-based” method scores each drug based on these
scores and distances between drugs and genes to rank given drugs. The
“RWR” method propagates these scores via a random-walk with restart
framework to rank the drugs.
The wrapper function score_drugs() can be used to score
and rank drugs for an individual tumor sample via the “distance-based”
or “RWR” method. The required inputs are:
-
driveR_res: data frame of driveR results. Details on how to obtaindriveRoutput are provided in this vignette -
drug_interactions_df: data frame of drug-gene interactions (defaults to interactions from DGIdb expert-curated sources) -
W_mat: (symmetric) adjacency matrix for the PIN (defaults to STRING v11.5 interactions with combined score > .4) -
method: scoring method (one of “distance-based” or “RWR”)
Example Input
In this vignette, driveR results for a lung adenocarcinoma case,
example_driveR_res, is used as the example input dataset.
Details on how to obtain driveR output are provided in this
vignette.
head(example_driveR_res)
#> gene_symbol driverness_prob prediction
#> 1 IFNA7 0.18945527 non-driver
#> 2 TJP1 0.13935021 non-driver
#> 3 IL20RA 0.09266710 non-driver
#> 4 SOST 0.08017132 non-driver
#> 5 NCOA2 0.06734062 non-driver
#> 6 ARAF 0.06232757 non-driver“Distance-based” Method
For this method, the score between a drug, d, and an altered gene, g, is defined as:
\[score(g, d) = \frac{1}{(dist(g,d) +
1)^2} driver\_prob_g\] where \(dist(g,d)\) is the distance between g and d
within the PIN, and \(driver\_prob_g\)
is the driverness probability obtained from driveR.
The final score for a drug is calculated as the average of the scores between each altered gene and d:
\[score(d) = \sum_{g \in G} \frac{1}{|G|} score(g,d)\] where G is the set of all altered genes.
Scoring of drugs for the example lung adenocarcinoma case
example_driveR_res via this “distance-based” method can be
performed as follows:
example_scores_dist <- score_drugs(example_driveR_res, method = "distance-based")This scores and ranks drugs via the “distance-based” method using drug-gene interactions from DGIdb expert-curated sources and the STRING v11.5 PIN with combined score > 700. Below, top 10 drugs are printed:
head(example_scores_dist, 10)
#> EVEROLIMUS PEMBROLIZUMAB CISPLATIN LETROZOLE CETUXIMAB
#> 0.009405251 0.008960906 0.008960906 0.008900282 0.008755715
#> BICALUTAMIDE SORAFENIB ENCORAFENIB REGORAFENIB TRAMETINIB
#> 0.008755715 0.008523904 0.008523904 0.008523904 0.008523904“RWR” Method
For this method, a random-walk with restart framework is used to propagate the driverness probabilities. A drug’s final score is its final propagation score.
Scoring of drugs for the example lung adenocarcinoma case
example_driveR_res via this “RWR” method can be performed
as follows:
example_scores_RWR <- score_drugs(example_driveR_res, method = "RWR")This scores and ranks drugs via the “RWR” method using drug-gene interactions from DGIdb expert-curated sources and the STRING v11.5 PIN with combined score > 700. Below, top 10 drugs are printed:
head(example_scores_RWR, 10)
#> BLOSOZUMAB;ROMOSOZUMAB TIAGABINE HYDROCHLORIDE
#> 0.0008476627 0.0004770740
#> KRN-330 SERELAXIN
#> 0.0004326723 0.0003804546
#> TIAGABINE LY-3009120;MLN-2480;XL-281
#> 0.0003554924 0.0002520392
#> OCRIPLASMIN ENCORAFENIB
#> 0.0002386693 0.0002312123
#> COLLAGENASE CLOSTRIDIUM HISTOLYTICUM COCAINE
#> 0.0002245975 0.0002227435