Onboarding tutorial: practical use cases

This tutorial is accompanied by a content package on Jinko.ai which contains:

• A computational model, implemented from the work of Ma et al. (2021), named Ma et al - Preclinical QSP model TCE inside jinkō
• A knowledge model, which describes the phenomena implemented in the CM, named Ma's preclinical QSP model inside jinkō

Knowledge model

The Knowledge Model will be your friend for this tutorial. All phenomena are described in wording, as well as in equations with direct links to the Computational Model. Before diving into the trials and simulations, we strongly advise that you read at least the first part of the KM (Model Overview). It will help you get familiar with the model, its context of use, its inputs and its outputs. At any point during the tutorial, if you wish to learn more about a phenomenon, look up the associated paragraph in the Implementation details section.

You will notice that the Knowledge Model includes Renderers, which are directly linked to the Computational Model. Inside the Renderers, you will find an associated reference for each objects. Some of these references are external links to publications in peer-reviewed journals, and some are links to assertions made inside jinkō.

The tutorial is divided into three parts, of increasing depth and difficulty. These parts are intended to reflect the typical workflow that a QSP modeler would use to study a model using jinkō.

• Part 1 guides you in reproducing results that were published using the same model
• Part 2 deals with the design of pre-clinical experiments assisted by in silico trials
• Part 3 helps you create a responder analysis using receptor expression levels

Preamble: getting ready to run trials in jinkō

Before we guide you to launch your first simulations in jinkō, here is a brief overview of the typical workflow used to run a trial inside the platform. You will be guided through the different steps in the tutorial.

Set up your environment

1. Create a folder at the root of the project with your name or your initials
2. Create an empty Document in this folder

Workflow tips

• We advise that you use this document as a report. You can import and list all the project items you create along the way by simply drag-and-dropping their card into the document. This will ease the navigation between the different items. You will also be able to write down any personal comments or notes, and to tag other people for discussion.
• Add your name (or initials) to each and every project item you create during the tutorial. It will make them much easier to find using search tools
• Create a new subfolder for each Trial you are planning to create. This will help keep the project environment tidy and easy to navigate

Part 1: Reproducing published results

In this first part, you will get familiar with the model and launch your first simulations. Since the model is re-implemented from published results, we will validate this re-implementation by reproducing some figures from the article. One main benefit of in silico clinical trials, in theory, is the reproducibility of results. In reality, it is often difficult to obtain comprehensive information regarding the published models. Reproducing published results is a good opportunity to gain a deeper understanding of how a model works and what the presented outputs actually reveal.

Reproducing Figure 1

Fig. 1 from Ma et al. (2021)

How was this figure created?

This figure shows the predicted tumor growth curves against the data that was used to calibrate the growth rates. From the article, we can gather the following information regarding how these two panels were produced.

• 1e6 cancer cells are engrafted into the mice at day 0
• Human T-cells are co-engrafted with the tumor cells, with various ratios. It is referred to as the E:T ratio (effector to target)
  • Panel A: E:T ratio is 5:1
  • Panel B: E:T ratio is 1:1
• No treatment is injected (control arm calibration)
• Variability is explored on a few descriptors using a virtual population of 100 mice
• The green line represents the mean trajectory, the green region is given by the standard deviation and the purple region is given by the 95% confidence interval for simulation results

0. Setting up the environment

• Create a subfolder Part 1 - Figure 1 - <your name> in your own folder

1. Creating a Vpop

1. Create a new Vpop design:
   • associate it with the computational model Ma et al - Preclinical QSP model TCE
   • name it Figure 1 - Vpop Design - <your name>
   • add it to the folder <your name>/Part 1 - Figure 1 (should be selected by default)
2. You will notice that by default some parameters are used as descriptors for the variability. So far, you have nothing else to configure and you can hit Create a new Vpop, give it a name (for instance Figure 1 - Vpop - 100 - <your name>) and choose 100 patients. This will cover the inter-individual variations that are shown in the Figure.

2. Creating a Trial

1. Once your Vpop is created, it is automatically open. You can now click on Create a trial, and name it Figure 1 - Trial - <your name>.
2. Some elements are already filled inside this new trial (the CM, the trial duration and the Vpop), but it is missing outputs descriptors and protocol arms. 
3. In the first step (Main trial simulations options), you can already add the following outputs descriptors, by clicking on 'Select descriptors': 

• volTumor
• Tumor.CancerCell
• Tumor.TCell

3. Creating protocol arms

1. Moving on to the  Protocol arms step in the trial editor, click on New set of Protocol armsto create a new protocol.

   • name it Figure 1 - Protocol - <your name>
   • add it to the folder <your name>/Part 1 - Figure 1

2. We are now going to setup the two protocol arms that correspond to the two different panels from the figure. Click on Select descriptors and in the list that appears (by default it shows you the parameters that have the tag ProtocolSpecific inside the model), select:

   • nbInjCancerCell
   • nbInjTCell
   • nbCograftedTCell
   • cograftingTCells
   • treatmentRegimenCibis

   

3. From the description of these parameters and the information we gathered about the figure, try to figure out the values they need to be set for each arm.

▼ Solution

Parameter	Value for arm A	Value for arm B	Comment
nbInjCancerCell	1e6	1e6
nbInjTCell	0	0	This parameter is used in the case T-cell intra-venous injection
nbCograftedTCell	5e6	1e6	This parameter is used in the case of T-cell co-grafting with cancer cells
cograftingTCells	1	1	Set this parameter to 0 to use T-cell intra-venous injection, 1 to use T-cell co-grafting
treatmentRegimenCibis	0	0	Set this parameter to 0 to use control settings (no treatment)

4. Creating a Measure design / selecting Output descriptors

1. We are not going to use a Measure Design for this first step, you can skip it for this part of the tutorial. 

Quick tip

 

If you do not select any output descriptors nor any measure design, you will not be allowed to launch your trial, since it would not produce any output. This will prompt an error that says:

 

 The trial will not output any time series: please consider setting 'outputConstituents' in the solving options to something non empty.

 

5. Setting the solving options

1. This step can be skipped for this part of the tutorial. 

6. Launching the simulations and visualizing the results

1. You can now double check that everything is set up, and click on Launch trial simulation. While the trial is running, you can follow the progress in the monitoring tab.
2. Once your trial has run, you can click on Create visualization, and name it Figure 1 - Visualization
3. The timeseries of the Descriptors you selected can be visualized in the Timeseries tab
4. The descriptors you selected should be plotted by default. If not you can select them by hand in the left panel ('outputs -> select descriptors), and then click on Update results.
   • Both protocol arms are plotted on the same figure. You can disable one by clicking on the legend directly

Quick tips

If you realize you launched a Trial with the wrong settings, it is not a problem. However, you cannot edit a Trial once it has run. You cannot edit any project item associated to a Trial that has already run either (Vpop, Protocol Design or Measure Design). What you can do is duplicate it, adjust whatever needs to be adjusted, and launch it again.

Question

The two different arms we have created don't appear to produce different results. Can you understand why?

▼ Answer

The only difference between the two arms is the number of engrafted T-cells. However without treatment present in the model, the behavior of T-cells is very limited. In fact they do nothing except dying and exfiltrating outside of the tumor. You can visualize this phenomenon by plotting the evolution of Tumor.TCell (which corresponds to the amount of T-cells inside the tumor) in time.

7. Compile all the information before moving on

1. Navigate to your own folder and open the report you created earlier
2. Drag-and-drop all the cards of all the project items you just created (Vpop, Trial, Protocol design, Visualization).
3. Add any information you find relevant to the document and, if necessary, tag anyone you ask for comments or review.

Reproducing Figure 2

Let us now reproduce some more interesting results, and investigate the action of the treatment. We will focus on panels A and B from Figure 2, and we will also investigate other model outputs.

Fig. 2 from Ma et al. (2021)

How was this figure created?

This figure shows the time evolution of the number of T-cells inside the tumor, following co-engraftment of tumor cells and T-cells. Panel A is the control (no treatment) and Panel B corresponds to a single injection of 2 mg / kg antibody at day 4. In both cases the E:T ratio is 5:1.

0. Setting up the environment

1. Create a subfolder named Part 1 - Figure 2 - <your name> in your personal folder

1. Creating a Vpop

1. We will actually use the same exact Vpop as for the previous figure, so you don't have to create a new Vpop. You can move on to the next step

2. Creating a Trial

1. Create a new Trial, and name it Figure 2 - Trial - <your name>
2. In step 2 of the trial editor (Virtual population), you can select the vpop you created in the first part of this tutorial. 

3. Creating protocol arms

1. In step 3 of the editor, create a new set of protocol arms. 

2. You will need to define values for the following parameters, for the two scenario arms:

   • nbCograftedTCell
   • cograftingTCells
   • treatmentRegimenCibis
   • treatmentDoseCibis
   • timeInjCibisCentral

   Can you guess what value each parameter should take?

3. Associate the treatment arm with its control arm (which should be the only other arm you created) in the appropriate column

   

4. Creating a Measure design

1. In step 4 of the trial editor, click on New Measure Design to create a new measure. Name it Figure 2 - Measures - <your name>, and add it to the folder <your name>/Part 1 - Figure 2
2. In model outputs, select
   • Tumor.TCell
   • Tumor.CancerCell
   • Tumor.CancerCellDead
   • volTumor
   • volTumorCancerCell
   • Central.Cibis
   • Tumor.Cibis
3. Let us add a few scalar results
   • Maximal tumor growth in treatment vs. control: add a reduce function to volTumor, and choose max. Select the observation period to be between 0 and 20 days, then in Measure Type choose Ratio vs Control
   • Total treatment infiltration: add a reduce function to Tumor.Cibis and choose auc with an observation period of 0 to 20 days

Selecting output descriptors

• Note that you do not have to define a reduce function for each measure added to a Measure Design. However, selecting a descriptor in the measure design will ensure that you have access to the timeseries of this descriptor.
• If you only wish to visualize the timeseries of a certain descriptor, you may also select it in the Output descriptors section of the Trial editor directly

5. Setting the solving options

1. This step can be skipped for this tutorial. 

6. Launching the simulation and visualizing the results

1. Once you are confident everything is set up, you can launch the Trial. While the trial is running, you can follow the progress in the Patients monitoring tab.
2. When it is ready, you can create a visualization and name it Figure 2 - Visualization - <your name>
3. You can visualize the timeseries of all the output descriptors that were added to your measure design
4. The measures you created (with reduce functions) will be visualized in the Scalar results tab (you will need to select them in by clicking on Select output and searching for the label you chose)

Can you observe the effect of the treatment?

Different phenomena are observed:

• Treatment distribution through PK mechanisms from the circulation to the tumor. The timeseries of the variable Central.Cibis shows the PK-profile as typically measured in vivo. The infiltration of the treatment into the tumor is evidenced by the scalar result you defined on Tumor.Cibis.
• T-cell proliferation, following complex formation at the interface with cancer cells. This is evidenced by the evolution of the timeseries of Tumor.TCell
• Tumor growth inhibition. You have access to various indicators of tumor growth including the scalar result giving the maximal tumor volume (as compared to the control arm) as well as the timeseries of volTumor, Tumor.CancerCell and Tumor.CancerCellDead. In particular, you can investigate the composition of the tumor in time.

7. Compile all the information before moving on

1. In your report, list all the different project items you have created and type your notes.

Interlude

Message from a clinician

After reading your report and observing the results from Part 1, a clinician has questions for you. It seems that in the results reproducing Figure 2, the cancer cells are almost completely eradicated after the treatment injection (indeed Tumor.CancerCell goes to 0).

Then how come the tumor volume is not 0? What process is driving the tumor reduction after all the cancer cells are killed? Given these results, does it make sense to investigate doses larger than 2 mg / kg?

Part 2: Treatment dosing

We will now investigate in detail the effect of the treatment and in particular the sensitivity to dose and to administration regimen. Feel free to come back to Part 1 to be reminded of the workflow and the elements that need to be defined in order to launch a trial.

Situation

Before running a pre-clinical trial, it is necessary to identify the most relevant treatment regimen and dosing range. A dose is relevant if it shows treatment efficacy, and if it produces minimal toxicity (which we can approximate by cytokine release in the blood). Use the following parameters in order to investigate the effect of dose:

• Treatment dose: this is the main parameter you want to vary, with doses ranging from 1e-4 mg/kg to 10 mg/kg (as well as 0 for the control arm). The corresponding parameter in the model is treatmentDoseCibis. We advise that you create less than 10 different arms, in order to make the visualization easier.

• Experiment design: a lot of different parameters are available to define the pre-clinical experiment. In order to keep things constrained, use the following protocol for all arms:

  Parameter	Value	Comment
  nbCograftedTCell	5e6	Use an E:T ratio of 5:1 (by default the amount of injected cancer cells is 1e6)
  cograftingTCells	1	Use the cografting method
  treatmentRegimenCibis	2	Repeated treatment injections
  timeInjCibisCentral	4	Start the treatment at day 4
  timeRecurrenceCibisTreatment	3.5	Each injection is separated by 3.5 days (bi-weekly regimen)

• Virtual population: you can use the virtual population created in the previous part, as it already describes the inter-individual variability

Outputs

Create a Trial which allows the visualization of the following outcomes:

• Tumor growth inhibition, measured by the tumor volume (volTumor or volTumorCancerCell, you can decide which one seems most relevant)
• Tumor cell lysis, which you can measure from the number of living and dead cancer cells in side the tumor (Tumor.CancerCell and Tumor.CancerCellDead). It might be interesting to compare this with the control arm
• Toxicity: a surrogate for toxicity could be to measure the area under the curve of cytokine levels (Central.Interleukin6 and Central.IfnGamma)
• In vivo binding efficiency: one main benefit of using in silico trials is the ability to track all outcomes, even when they are not observable in vivo. In this case, the model allows to visualize the amount of antibody that successfully binds to its targets. You will find in the KM a full paragraph describing this phenomenon under Ternary Complex formation. The corresponding output descriptor of interest would be SynapseTce.CeaCeaCibisCd3
• T-cell proliferation inside the tumor, as measured by Tumor.TCell

Question

What would be the best pre-clinical experiment design in order to assess the efficacy of this treatment? In particular:

• What is the minimal dose for which the treatment is efficient (for a given regimen)?
• What is the maximal dose after which the effect of the treatment is saturated?
• How does the toxicity evolve with the dose?

Further investigations

After this preliminary dose effect study, you can choose to investigate other design parameters, such as the treatment regimen, the treatment starting day and recurrence or the E:T ratio.

Interlude

Message to the clinician

After analyzing the results from the dosing investigation for pre-clinical experiments, you may write down your comments and observations in your report. Then tag your trialist collaborator in the comments and ask for their opinion regarding the identified relevant dose range. In particular, you should now be able to answer with more detail the question that was asked previously:

Does it make sense to investigate doses larger than 2 mg / kg?

Part 3: Responder analysis

Situation

Having identified a potential treatment as well as a trial design, we can investigate the selection of patients. In our case, one main parameter that characterizes the patients is the expression of the target cell receptor, CEA. In particular, some mutations may impact the expression of this surface receptor. We would like to identify the impact of these mutations on the treatment efficacy, using model simulations.

Instructions

1. In order to investigate the effect of CEA expression levels, create a Vpop where you specify a distribution for totalCeaCancerCell, which represents the number of receptors per target cell. To give you an idea, this number ranges from approximately 1 to 1e6, but you may want to refine this range once you obtain some simulation results.
2. You may use the protocol identified in the previous part, which ensures that the treatment is efficient (for a patient with an average CEA expression level)
3. Using jinkō's visualization tools, investigate the effect of CEA expression level on treatment efficacy

Quick tips

• In a visualization, you can define custom groupings and filters on the patients

• In particular, you can group patients based on the value of a specific parameter. In this case you could create groups based on the value of totalCeaCancerCell. Note that the grouping in separate bins will separate the parameter values using a uniform distribution. This may create unbalanced groups if the prior distribution of the parameter is log-uniform.

  

• You can also choose to exclude some patients from the visualization by using filters. For instance, if you do not need to visualize the control arm at all, you can add a filter on the treatmentRegimenCibis parameter value

• If you only need to filter scenario arms from a single plot, you can also click on their names in the legend

Identifying responders

Can you identify a threshold of CEA expression (amount of receptors per cell) which separates responders and non-responders?

There are different ways to defined responders. We suggest the following criteria, feel free to use any or all of them. In each case, the numerical threshold will depend on the treatment regimen you are using for this analysis.

• The tumor volume at day 20 is lower than a certain threshold
• The maximal number of cancer cells between day 0 and 20 is lower than a certain threshold
• The ratio of maximal tumor volume against the control arm is lower than a certain threshold