1st talk
Speaker: Megan Gibbs (AstraZeneca)
Title: Impact of Big Data and Artificial Intelligence on Drug Discovery and Development
Abstract: Artificial Intelligence holds the potential of driving innovation, automation and acceleration of drug development. This presentation will describe the evolution of modeling and simulation approaches including artificial intelligence throughout the drug discovery and development continuum. The AstraZeneca approach of decreasing attrition using the 5R’s will lay the foundation for where AI approaches could benefit the drug development process. Specific case studies in both the preclinical and clinical areas will converge to highlight the translational prediction of therapeutic index with the theory that an earlier understanding of therapeutic index will increase probability of success, decrease cycle time and increase innovation in drug development.
2nd talk
Speaker: Gift Nyamundanda (GSK)
Title: An evaluation of the utility of Machine Learning in identifying biomarkers predictive of longer term clinical efficacy using longitudinal data from multiple Ph III trials
Abstract:
To accelerate the clinical development of molecules we were interested in identifying surrogate biomarkers that can be used to predict long term clinical efficacy. Such biomarkers might allow us to run shorter or smaller Phase II clinical trials.
Exploratory biomarker analysis using several machine learning (ML) algorithms was carried out using longitudinal data from five Phase III trials. Models that can be used as early predictors of long-term clinical efficacy were derived and their predictive performance compared. The data consists of more than 3,500 patients (who received active treatment or placebo) from five Phase III clinical trials. Twenty-seven clinical laboratory markers were measured at baseline, week-8, week-24, and week-52 in each trial. The clinical outcome of interest is a continuous score at week-52 measuring disease activity in patients.
We will share results from this analysis and discuss signatures for response prediction derived using different ML approaches, some methods assuming time independence (ignoring correlation in observations over time) and others capturing time trends in the data.
3rd talk
Speaker: Paul Newcombe (GSK)
Title: Causal Machine Learning for Biomarker Subgroup Discovery in Randomised Trials
Abstract:
Decreasing costs of high-throughput ‘omics, as well as new technologies such as the Olink proteomics platform, has driven wider application in clinical trials, for example to inform precision medicine strategies. However, data-driven characterisation of patient subgroups with enhanced (or weaker) treatment effect remains a challenging problem, particularly when searching over high-dimensional biomarkers. With growing recognition that traditional approaches (e.g. exhaustive biomarker-treatment interaction testing) are sub-optimal, several promising methods have recently emerged that combine machine learning tools with concepts from causal inference. In principle, they offer greater power through less conservative multiplicity control, and the ability to capture complex multivariate signatures which may be missed during one-at-a-time testing.
I will describe three causal machine learning methods for responder subgroup detection; the “Modified covariate Lasso”1, “Causal Forests”2, and the “X-Learner”3. I will compare and assess their performance in a modest simulation study motivated by real biomarker trial datasets being generated within GSK. I will then share some early (gene-anonymised) results from an on-going application of these methods to detect and predict responder subgroups from transcriptomic data measured in two Phase 3 Lupus trials. I will close with a discussion on the benefits, and limitations, that we found with existing methods in this space.