Immunetrics

Our Platform

Building reliable mathematical models is a complex process with several stages, each requiring unique computational demands. The Immunetrics platform addresses routine problems by automating error-prone and laborious tasks, which frees the modelers to focus on the science of modeling. The Immunetrics tools also exploit novel computational techniques developed by Immunetrics scientists to offer insight into the hard questions faced by both developers and end-users of these models. Applying this two-pronged approach at every phase of a project, the Immunetrics platform helps maximize the success of modeling endeavors.

Model development

At the core of our computations lies an efficient "unified simulation engine" shared by each component of the platform. This engine is designed to utilize distributed computing resources for fast, scalable execution. Immunetrics employs sophisticated numerical integrators for simulation to avoid the stiffness problems arising from multiple time scales, common in biological models. This engine also supports user-specified rules, applied when desired conditions are met – for example, changing treatment regimen based on patient status.

Model analysis and visualization

Throughout the development process, modelers need convenient ways to modify the structure of their models, solve them quickly and accurately, and visualize the model output. The Immunetrics platform frees modelers from the burden of writing low-level code to implement models. Instead, users write models in an easy-to-read, natural equation syntax, which is automatically converted to efficient, compiled code for maximum performance. Model editing is integrated with a cross-platform graphical user interface (GUI) that allows users to set conditions, simulate, and manipulate plots of time course and target data.

Model analysis is also crucial for debugging, comparing possible implementations, assuring parsimony of models, and understanding the limitations of models. Our platform offers tools for dealing with many of these problems, including sensitivity analysis, prediction intervals, etc.

Optimization

Numerical optimization is key to estimating unknown model parameters and optimizing treatment scenarios or selection criteria in simulated clinical trials. To this end, Immunetrics has built an extensive library of powerful, nonlinear optimization algorithms. These codes are customized and tuned for the problems of fitting biological data: noisy or missing data, data of drastically different scales, balancing competing scenarios, and bounding of parameters. In the absence of quantitative data, these optimizers can also accept qualitative heuristics as constraints. Tools to help characterize large scale optimizations include computation of correlation matrices, confidence intervals for estimated parameters, clustering of model ensembles, and model simplification to remove dependencies. Large-scale models such as ours often present the challenge of many unknown parameters, and even sophisticated estimation algorithms can only go so far in providing meaningful parameter estimates. The Immunetrics team has significant experience in tackling parameter estimation in highly under-constrained scenarios, while still making model predictions meaningful and quantitative.

Clinical trial simulation

Simulating clinical trials is a natural application of mechanistic models. Immunetrics's platform can scalably simulate very large populations in a distributed and timely manner. Because defining realistically representative populations is critical, users have extensive control over the distributions of variations imposed on generated cohorts.

Although we have developed this platform for in-house model building, it can be made available for use by our customers. Further, we are able to translate our models into a discrete unit that our partners could then run in MATLAB or a similar application.