Modeling Disease Progression in Mild Cognitive Impairment and Alzheimer's Disease with Digital Twins

Abstract

Alzheimer's Disease (AD) is a neurodegenerative disease that affects subjects in a broad range of severity and is assessed in clinical trials with multiple cognitive and functional instruments. As clinical trials in AD increasingly focus on earlier stages of the disease, especially Mild Cognitive Impairment (MCI), the ability to model subject outcomes across the disease spectrum is extremely important. We use unsupervised machine learning models called Conditional Restricted Boltzmann Machines (CRBMs) to create Digital Twins of AD subjects. Digital Twins are simulated clinical records that share baseline data with actual subjects and comprehensively model their outcomes under standard-of-care. The CRBMs are trained on a large set of records from subjects in observational studies and the placebo arms of clinical trials across the AD spectrum. These data exhibit a challenging, but common, patchwork of measured and missing observations across subjects in the dataset, and we present a novel model architecture designed to learn effectively from it. We evaluate performance against a held-out test dataset and show how Digital Twins simultaneously capture the progression of a number of key endpoints in clinical trials across a broad spectrum of disease severity, including MCI and mild-to-moderate AD.

Interpretable machine learning for high-dimensional trajectories of aging health

“…We evaluate these aging trajectories by comparing with the observed test population. We train a logistic regression classifier to evaluate if the synthetic and observed populations can be distinguished [18,19,42,43]. We find that this classifier has below a 57% accuracy for the first 14 years past baseline (Supplemental Fig.…”

“…Recently, other machine learning models of aging or aging-related disease progression have been emerging [12,[18][19][20]43]. Since they each differ significantly in terms of both the datasets, types of data used, and scientific goals, it is still too early to see which approaches are best -and for which data and what goals.…”

“…To assess the quality of our synthetic population, we train a logistic regression classifier and evaluate its ability to differentiate between the observed and synthetic populations [18,19,42,43]. Ideally, a synthetic population would be indistinguishable from the observed population, giving a classification accuracy of 50%.…”

Artificial Intelligence for In Silico Clinical Trials: A Review

“…In silico trials are an ongoing research area aimed at using simulations or machine models to replicate some, if not all, parts of a regular clinical trial. [12] Conditional Restricted Boltzmann Machines Bertolini et al [13] Conditional Restricted Boltzmann Machines Choi et al [14] Generative Adversarial Networks Guan et al [15] Generative Adversarial Networks Cui et al [16] Generative Adversarial Networks Baowaly et al [17] Wasserstein Generative Adversarial Networks Zhang et al [18] Wasserstein Generative Adversarial Networks Biswal et al [19] Variational Auto-encoders Lee et al [20] Auto-encoders and Generative Adversarial Networks Xu et al [21] Conditional Generative Adversarial Networks Biosimulation Eddy and Schlessinger [22] Differential Equations Kovatchev et al [23] Differential Equations Man et al [24] Differential Equations Gillette et al [25] Differential Equations Compte et al [26] Differential Equations Baillargeon et al [27] Differential Equations Zhang et al [28] 3D Printing Low et al [29] 3D Printing Herland et al [30] 3D Printing Yu et al [31] Reinforcement Learning Individualized Predictive Modeling…”

“…They showed that digital subjects generated by the CRBM were statistically indistinguishable from actual subjects enrolled in the placebo arms of clinical trials for MS. Similarly, Bertolini et al [12] described a generative model composed of two CRBMs that accurately simulated subjects with Mild Cognitive Impairment (MCI) and Alzheimer's Disease (AD). They showed that linear classifiers were unable to distinguish actual subjects from their digital twins based on the generated features; Bertolini et al…”