Machine-learning and digital-twins for rapid evaluation and design of injected vaccine immune-system responses

Abstract

A computational framework is developed that researchers in the field can easily implement and subsequently use as an efficient tool to study the immune-response to a vaccine injection. There are three main components to this work: Part I-Digital-twin construction: An approach is developed that efficiently simulates the time-transient proliferation of cells/antibodies (proteins) and regulator/antigens (deactivated toxin) to an injected vaccine within tissue possessing complex heterogeneous microstructure. Here, we use the terms “cells” and “antibodies”, as well as “regulator” and “antigen” interchangeably. The approach utilizes two strongly-coupled conservation laws: (a) Conservation Law 1: comprises (a) rate of change of cells/antibodies, (b) cellular/antibody migration, (c) cellular/antibody proliferation controlled by a cell/antibody mitosis regulating chemical (antigen), (d) cell/antibody apoptosis and (b) Conservation Law 2: comprises (a) rate of change of the cell/antibody mitosis chemical regulator/antigen, (b) regulator/antigen diffusion, (c) regulator production by cells/antibody and (d) regulator/antigen decay. Part II-Efficient computation: A technique based on a voxel (3D “volume pixels”) representation of tissue microstructures and corresponding digital solution methods is developed for the calculations, which avoids computationally expensive steps involved in usual Finite Element procedures such as topologically conforming meshing, mapping, volume integration, stiffness matrix generation and matrix-based solution methods. The process proceeds by converting the tissue microstructure into voxels. The problem then becomes “digital” on a regular “voxel-grid”, directly manipulating voxel values, allowing extremely fast methods to be used to construct derivatives and to iteratively solve the system with minimal memory requirements. Part III-Machine-learning: The rapid and efficient computation allows for many vaccines to be tested quickly and uses a genomic-based Machine-Learning Algorithm to optimize the system. This is particularly useful for rapid design of next-generation vaccines and boosters for disease strain mutations. Numerical examples are provided to illustrate the results, with the overall goal being to provide a computational framework to rapidly design and deploy a vaccine for a targeted response.