Hi, I'm Adrian. In this demo, Jules OS answers a question about the industry's top-selling drugs. So, let's get started.
On the landing page of Jules OS, the user can enter the query, what is the revenue of the 10 top-selling drugs that target proteins highly expressed in the liver? Jules OS immediately gets to work, formulating a plan. Please note, for this demo, inference times have been sped up. Once this process is complete, Jools OS describes its plan and the reasoning behind it.
On the right-hand side of the screen, the sequential steps of this plan are shown in a mini-map. Below the plan, Jools OS asks me if I would like to continue to the first step. I respond, yes, to move forward.
Next, JulesOS begins executing the first step to get me all proteins highly expressed in the liver. Results are streamed back in real time and I can see the progression on the minimap. Here I can see that JulesOS decides to retrieve the genes dataset then designs a SQL query to extract the relevant data.
You can see that JulesOS is filtering for proteins expressed in the liver. A preview of this extracted data. is also provided.
Since this looks good, I'm going to click to proceed to the next step, which is get me the drug protein interactions for all drugs. In this case, Jools OS retrieves drug data and again designs a SQL query to extract the data. This is successful, so I again proceed to the next step of the plan, which is get me sales information for all drugs.
By inspecting the minimap, I can see that this will be the last data extraction step to complete. This also looks good. Therefore, I will click to proceed to the final step. Now this is the big one, where the magic happens.
Here, Jools OS analyzes and reasons about the extracted data to answer my initial question. It writes and executes its own Python code, debugging it as necessary along the way. I can see that Jools OS starts by breaking this data wrangling exercise down into a series of sub-steps. The first sub-step is to merge the extracted gene data with the extracted drug data.
Jools OS provides a sample of the output data for our review. The results look good. Let's continue. The next sub-step is to merge the extracted sales data. On first try, this join resulted in an empty data frame.
indicating that the approach failed. This can sometimes happen due to naming conventions, so I asked to normalize the names. All uppercase letters are converted into lowercase, and any leading or trailing white space is removed.
Looking good? Let's continue. The final step sorts the table.
by the total reported sales in 2022 and returns the top 10. This looks okay, but I want to ensure target protein names are displayed alongside each drug and the associated sales data. You can see the plan is updated on the fly. I click yes.
to proceed with the plan. Finally, we get to our answer. Jules OS has created a list of the top 10 selling drugs that target proteins highly expressed in the liver. We have a well-reasoned, data-driven answer to our question. I want to emphasize that this result was automatically generated without me needing to know about GSK's data ecosystem.
or having the ability to code. We believe this tool has real potential for augmenting the abilities of our scientists to make a step change in research productivity at GSK. Thank you for watching this demo.