Jeremy Sohn, VP, Digital BD&L, Novartis, presenting on Nov 5, 2020. Title: "A New era for Life Sciences: Reimagining medicine with data and digital". Presentation begins with using data and digital to diagnose cancer. Overview of the "Map of Life", an initiative to explore, and analyze Novartis' historical clinical trial data, with applications like Qubit, Chord, Chronograph. There is an example of taking the first step toward in-silico development.
UC goals:
Learn from the past: leverage experience from stability data generated from 1000s past formulations
Connect heterogeneous data: integrate data in the same context coming from diverse source systems
“Bring your data and models”: make your data and models available to a wider user community
User journey: problem statement #1
I am testing two variants of my drug product, which is currently formulated as a liquid in vial. I am exploring two possible strengths of this drug: 5mg and 10mg. I have one batch of each strength available for testing and I am putting these on stability testing. My goal is to validate if this drug product is stable chemically and physically both at 5mg potency and in 10 mg potency. Performing the experiment, my drug product resulted not stable at certain conditions (e.g. 6 months at 40°C/75% RH)
Opportunities for Data Science and AI::
Is the failure at 6 months a true failure, or a data outlier? Should I try to continue or should I stop here and change formulation?
The failure is at 40°C/75%RH, but how critical it is for 25°C/60% RH?
Could we have predicted the failure basing on historical data? If yes, what is the common element to the failing experiments? Could we have started mitigation work earlier?
Can we help finding a root cause? Were there similar failures in the past? Were there anomalies in the production?