Integrating machine learning into the in silico clinical trial pipeline

In silico clinical trials offer a powerful tool for overcoming several limitations of traditional clinical trials. Conventional trials are time- and resource-intensive, typically designed to assess average effects across a population while being restricted to studying the impact of a fixed treatment protocol. In contrast, in silico trials are cost-effective, flexible in their design, and able to explore heterogeneity in treatment response. These trials generally rely on expert-developed and data-calibrated mechanistic mathematical models and the identification of model parameterizations that satisfy biological or clinical constraints. With the growing availability of multi-scale and high-resolution clinical data, it is the opportune time to thoughtfully consider how machine learning (ML) methods can enhance the feasibility, interpretability, and reliability of these in silico trials. In this perspective piece, we explore both the opportunities and the challenges of introducing ML tools at various stages of this process, from biomarker identification to interpreting the results of the trial. We argue that in the hands of an expert modeler, the thoughtful application of ML tools can result in more accurate and informative in silico clinical trials that may potentially accelerate drug development and find the right drug/protocol for the right patient.