Figure 1
Flowchart and summary of our approach.
(a) Flowchart of the analysis. Our study is divided in three steps: i) experimental setup and data collection; ii) training of models/algorithms; iii) prediction. Experimental setup and data collection: In the experiments, pig blood circulates from the animal to a perfusion chamber (Badimon Chamber) containing one of the three different vascular tissues considered triggering thrombi (tunica media, pig tendon, subendothelium). We collected platelet deposition counts for different experimental conditions such as perfusion time or shear rate (see Table 1 and Methods). We performed experiments with four different animals. Training: We consider all the collected input (experimental conditions) and corresponding platelet deposition data for three pigs. With this information we train the models/algorithms to get a good agreement between model/algorithm outputs and known platelet deposition values. Prediction: We now consider the data collected for the remaining pig. We use the experimental conditions in that dataset as inputs to the trained model/algorithm to obtain predictions of platelet deposition values for each set of conditions. We test the prediction power of each model/algorithm by comparing predicted platelet deposition values to measured platelet deposition values. We carry out steps ii) and iii) for the four different combinations of training (3 pigs) and test (1 pig) datasets. (b) Advantages and limitations of each of the computational approaches for platelet deposition prediction that we consider in our study: a mass-transfer boundary layer model, the Random Forest algorithm and a phenomenological model (see text).
