P53 Target Identification

Selection of targets potentially involved in regulation of P53 expression.

In this section, we will make 2 different approaches. 1st one is a simple correlation analysis between the actual and predicted P53 score and the  experimental activity of the molecule at each specific target. The P53 score is calculated aggregating data by molecule-target pairs in a variable called Avg(combinedP53PotencyValue) composed of existing experimental values and predicted values when there is  no experimental record. The potency of each molecule at the specific targets is represented by Avg(chemblActivityScoreC). The corresponding correlation coefficient between the two variables and the slope are calculated and used to select what activities at specific targets are more related to P53 activity. Then, a second selection based on potency at P53 (potency at target could be used as well) is applied to highlight most interesting targets (see dashboard below).

Correlation coefficient vs slope plot (left). Average potency at P53 vs number of events (right). Dots marked in the left chart appear in the right chart to produce correlation plots coming next.

Once selected best correlating targets with higher average potency at P53 expression assay, the correlation plots by target can be visualized. Next three charts show the P53 to target correlation plots for all sort of assays, protein based assays, and cell-based assays.

P53-target correlation plot. All sort of targets.

P53-target correlation plot. Protein targets.

P53-target correlation plot. Cellular targets.

Table containing 40 best correlating assays.

Second approach joines P53-active molecule-target pairs into single molecular targets, calculating different P53 scores based on the results obtained with different predictive algorithms, and a target score, which is the result of the average potency at target of the P53 active compounds (actual and predicted).

When the P53 active molecule-target pairs are aggregated by target, a plot of the best scored targets can also be obtained to select protein targets with higher probability of inducing P53 expression:

Target score vs number of records per target.

Charts below shows the target selection by target score (bottom-right) with the values aggregated by target for the decision trees rate of predicted activity (countRatioRPart), random forests binomial (countRatioRF) and random forests regression (combinedPredictionScore).