Digital Twins In Drug Discovery, Development, And Personalized Therapeutics

Abstract

DTs are well-established in the fields of aerospace and manufacturing engineering; however, their application in drug discovery and healthcare is rather poorly explored and lacks consensus on the definition. The present review synthesizes information from various sources throughout the stages of drug discovery and highlights the difference between the dynamic evolving digital patient twin (DPT) and the traditional predictive static model. The present review utilizes the classic digital twin architecture (physical, virtual parts, and data interface), which is applied to the DPT framework. Moreover, the DPT framework is extended by the examples of molecular, cellular, and organ on chip models, virtual patient cohorts, and simulated clinical trials. The scope of generative AI technologies, including generative adversarial network (GAN), variational autoencoder, transformer, restricted Boltzmann machine, and graph neural network, is included. Generative AI distinguishes genuine DT from conventional approaches owing to the fact that it provides for continuous and personalized simulation of a patient’s health path. Case studies regarding immune system conditions, rare diseases, oncology, and additive manufacturing are discussed alongside regulatory documents of the FDA and EMA. Regulators have already approved certain pieces of evidence related to DT, and such initiatives as DITTO and I-SPY 2 demonstrate improved treatment matching and more effective trials. Nonetheless, overall there has been some progress, albeit uneven. The vast majority of DT implementation cases are at an initial stage due to limited data, low interpretability, and validation issues.