10106925SHARPEN: SHARing data to accelerate Pharmaceutical manufacturing Efficiency across trusted Networks; A Framework for Risk Assessing the Value of Federated Learning to Improve the Fidelity of Models in Pharmaceutical ManufacturingClosedCollaborative R&DInnovate UKThe Pharmaceutical industry is really good at employing molecular-level chemistry models to help predict likely new cures for diseases. On the flip side it is bad (vs. other industries) at applying modelling to manufacturing to predict how and when to make products most effectively. This is for a number of reasons, one of which is the lack of good quality data. This would enable the models to make better predictions; as more good data, leads to better predictions. One way to get more data is to make it yourself, but that's expensive and wasteful as a solo effort. It's much better if you can share. Companies find it hard to trust each other sharing data though, as they are competitors. So sharing is blocked by cyber security concerns, commercial threats, and the lack of certainty that the data will be used as intended. One way to fix these concerns is to employ Federated Data sharing technologies. These novel digital tools address the concerns of 'who has access to data' and 'why', because you can control these aspects centrally. They are also very cyber secure. They do not solve the concerns of commercial threat, however. As, if you share all of your data, you may well give away valuable secrets. The obvious solution is to share data (through the new technologies) but share segments of the data, not the whole. This way modelling outcomes can be achieved more effectively, but you're not giving away valuable information. The trouble here is that there is little evidence that redacted datasets lead to better modelling outcomes. There is also a business risk, as there are very few practical tools available to determine how much data is 'too much' data shared. This project (SHARPEN) intends to deliver a platform for data sharing (so we can assess it) that runs across R&D data to manufacturing (ensuring good data transfer across all relevant data) and deliver a risk assessment tool (to enable rapid assessment and subsequent sharing of data), as well as working out what someone would pay for that service. We will deliver the outcomes through diverse partners who have significant experience in the pharmaceutical sector and who've successfully worked together in the past. We will enable a number of market ready digital tools in the process. Ensuring medicines manufacturing becomes more efficient through effective use of models to accurately predict what to do next.