The case study data is indeed proprietary and confidential. What we can disclose is given in answers to the questions below (especially the following question). In the same way, your case data will be proprietary and confidential and unique to you.Read more about inSee Question about inSee: Can you tell us what sample sizes you were using for predicting 10-15 years ahead?
There are three datasets: 1.) Learning/Evaluation Dataset (LED). ~1000 examples of ‘successful’ and ‘unsuccessful’ innovations, used to train and test individual modules in the pipeline prediction model. 2.) Test Dataset. 50 examples of ‘successful’ and ‘unsuccessful’ medical breakthroughs, assembled by client Chief Medical Officers. Success means treatment (usually a drug) received FDA approval and then had significant impact on mortality. Unsuccessful means failed at some point in the development process (for example, failed at Phase III clinical trial, or was granted patents and generated excitement but failed to enter clinical trials). This is the main dataset used by the client to evaluate the prediction model, as the success/failure labels for the examples are directly relevant to their business. 3.) Operational Dataset: ~10,000 examples, all unlabelled (i.e., not known yet if they will be successful). Our model identified ~10 as very promising, and client CMOs evaluated these and were very impressed with the predictions (it was agreed all were very promising, and about half were unknown to the CMOs).Read more about inSee Question about inSee: What were % the false positives and false negatives?
The %FP and %FN are ~5% each (matching the overall class-average accuracy) for the Test Dataset – this is the main dataset used by the client for evaluation. Accuracy on the LED is higher (and so %FP and %FN are lower). Accuracy on Operational Dataset is of course not known. The balance between FP and FN rates can be adjusted, and we recommend that when deployed the learning algorithm should penalize FN more than FP in this application.Read more about inSee Question about inSee: If someone decides to stop development of a line of science behind a firewall, how do you see that?
We do NOT access anything apart from publicly available data, all our results are based on publicly available data. If people choose not to publish we cannot detect it.Read more about inSee Question about inSee: Why can you do this, and Watson cannot?
We do not try to do what Watson does. We looked at what Watson does, and it is not the strategy that we need to apply to do what creates the impactful outcomes that we are looking for.Read more about inSee Question about Volv: Can we set you a challenge?
Indeed, once we have an NDA and a Proof of Concept budget agreed. We recommend forming the challenge in a collaborative effort.Contact Us
What is your challenge? Want to be inspired? Contact me Christopher Rudolf Founder & CEO
+41 79 552 79 11