Sub-sampling of a virtual population, phase portrait, trial versioning and a beta version for model calibration!

During the few last months, we introduced significant enhancements in our modeling and simulation features, leading to more sophisticated exploration of trial outcomes. Key developments include the implementation of jinkō's virtual population sub-sampling, comprehensive trial versioning,  phase portrait and the preliminary phases of our new calibration module.

Sub-sampling: iterate on your simulations with refined virtual populations

Use outcomes of your trial simulations and jinkō’s embedded logic to identify subsets of patients with characteristics of interests, on both inputs and outputs of your models, and re-run simulations on these populations.

Once a trial is completed on an initial virtual population, you now have the ability to select scalar measures (typically used in your initial trial) to  apply constraints that will result in a new subset of patients. Jinkō’s algorithm will optimize the selection of patients, based on various options, in order to maximize likelihood between the subset of patients and the constraints. Constraints can for instance be distributions or Kaplan-Mayer curves, used for survival analysis.  Note that, in the same context, you can also add a filter to completely remove patients with some characteristics.

For example, you may run a simulation with a viral infection on a virtual population where some input parameters vary: 

• In this context, the available literature gives information on the distribution of the level of seroprotection (HI titers) during infection. Subsampling allows to select a subset of patients among the Vpop that best matches this distribution in the simulated trial. 

• Moreover, since you are likely to be interested in the symptomatic patients from the placebo arm, you can use a filter to remove all patients with a max viral load lower than the level of detection.

Phase portrait: visualize your system’s dynamics

A phase portrait offers a visual representation of the trajectories of a dynamical system within a multidimensional phase space. Observing these trajectories allows for the inference of the system's dynamics, such as stability, chaos, or periodicity.

Here is an example below for the Lorenz Model over a simulation period of 30 seconds, set up with three different initial conditions, where we can see a butterfly-shaped pattern.

Learn more about how to use it in jinkō with your model here.

New trial versioning capabilities: easily and safely apply variations in your trials

The trial editor is now fully versioned in jinkō, enabling you to make as many variations as you like to a trial, for instance as we have seen above re-running it with a sub-sampled population, then observing results and iterating again. The versioning panel allows you to quickly navigate between the different versions, identify the ones with results generated, and restore a version if needed, all in one place.  

Model calibration: the beta version of our new module is out! 

Calibration is a process in which a priori unknown and estimated input descriptors of a computational model are refined through a numerical procedure, more specifically optimization routines in order to achieve a desired behavior of the model. This behavior is characterized by the Objective function to minimize.

This is a critical step when wanting to measure the coherence of your model in observation of samples of real-life data from clinical trials.

We are launching our first installment of the calibration module in jinkō, which allows the following (from a model): 

• Optimization of the fitness criteria: this is a metric, maximized by our calibration algorithm, of the match between outputs of the simulation and the expected behavior covered by a Scorings set and/or a data table used as inputs.

• Creation of variations in the simulation via a choice of outputs / measures and the configuration of protocol arms

• Options refining, such as inputs to calibrate, solving options, population size or completion conditions for the simulation. 

Learn more about calibration here.

Please note that this feature is at the moment only released for selected users, but  you can register to become a beta tester of the feature.

And more 

• Improved speed of solving for simulations: we went from 180 patients solved / second from our previous benchmark (for an average virtual patient's complexity)  to 400 patients / second.  For illustration, some recent simulations with more than a million patients could thus be entirely solved with jinkō in less than an hour.  
• Improved design of the minified representation (‘cards’) of items in the right-side appPanel. This redesign stemmed more particularly from the need for a better readability of references and extracts cards in the panel, in order to quickly identify them to drag and drop them in a document ; it has then be applied to all items cards that also benefit from it.