Let's come to the last section of our lecture. On strategies for integrated testing strategies. What is important to move forward? You just saw in the previous section that for skin sensitization, you can actually achieve productivity's which are very high, 90% of the substance were correctly predicted. Even could predict potency for skin sensitization. But how good do we have to be? I already said we are achieving values of reproducibility of the animal test. And what is this statement based on? We actually did some work more recently, which allows us to ask how good do we need to get our testing strategies. And for this purpose, the mind, the data of the UPN reach process. The large program in Europe taking place from 2000 and six legislation finalizing in 2018, which is aiming to register existing chemicals. We downloaded all of the publicly available data, which is about 9,800 chemicals in December 2014, and we found more than 800,000 chemical studies associated with these chemicals. So a lot of toxicological information worth billions of dollars of testing costs. And we made this machine readable. The respective publication is shown here. Actually the first of four publications which were put out in the same issue of ALTEX. So if you're interested in some more of the details, I refer you to this issue of ALTEX. And the fact that many of these substances actually been tested more than once, allowed us to really look into how reproducible is the testing for skin sensitization. You might have heard that several type of tests are being applied for testing whether substances can sensitize an animals. We do have the Buehler assay, and the guinea pig maximization test, the patch-test which are all done on guinea pigs, and we have the local lymph node assay, which is done on mice. Which is the preferred method nowadays, because it is less stressful to the animals. And what you can see here is how well the different tests do predict each other, or how well they predict themselves when repeating. So the diagonal shows that the Buehler assay for example is 95% reproduceable. The guinea pig maximization assay, 93%, or the patch-test is 92% reproduceable. The local lymph node assay which our point of reference is just reaching 88%. And this means that the productivity is the cell culture assays have achieved and shown in the previous section of the lecture, are exactly in the range of reproducibility of these animal tests. And they're even better than different animal tests predict each other. If you see that a Buehler assay for example predicts a local infant assay only with 77% probability, this is far less than the cell culture assays predicted the outcome. This shows you that by objectively studying how good the animal test is, we can define entry criteria when an alternative approach and integrate testing strategies good enough to actually replace the animal. Some very important thoughts on the future of integrated testing strategies originate from this workshop which was held in 2009, published by Kinsner-Ovaskainen and others which was held by the European Center for the Validation of Alternative Methods and European Partnership for Alternative Approaches to animal testing. And among the recommendations are some which we need to take into account for the future. One of them is important to take account, flexibility in combining different building blocks, components of ITS, that we need to identify ways of optimizing these things by identifying the minimal number of components and combinations to achieve a certain predictive capacities to be efficient. We don't want to do 18 tests for endocrine disruption. To call something estrogenic, endocrine disrupter. And we have to understand the applicability domain of single components and the whole integrated testing strategies. And at the end, it's also about the efficiency of this whole thing, and this is not only predictivity, it's cost time and technical difficulties as discussed earlier. But let's come back to the applicability aspect. The next slide shows you depending on how we combine logically different endpoints, we have different consequences for how applicable a testing strategy is. If we use Boolean algebra of a and b. So if a and b are positive tests or if a or b are positive tests, it will be only the overlap between the two which is actually allowing us to which we can apply the testing strategy. Same for, if we require that a not b is positive, and so on. And this shows you that we can start looking into how we combine assays not to lose too much above the chemistry. There's quite interesting opportunities also to combine these things proballistically of fuzzy logic, for scoring type of assays are helping us to take decisions where don't have the clear Boolean logic of a and b or b, not b type of decision and combinations. So you see that there's a quite a few of elements in optimizing testing strategies. Considering their applicability, domains, and how to maintain flexibility of these assays, and how to open up for the replacement of animal tests by actually comparing with the performance of the animal tests the performance qualities.