Can digital twins help identify treatments for preterm birth prevention?

Basic Applied Science
14 followers
5 Video Views·Nov 13, 2026
Digital twin models have the potential to simulate the body and pregnancy, helping to predict the risk of preterm birth, evaluate the effectiveness of treatments, and personalize interventions—thereby supporting safe and effective prevention of preterm birth.
AI Generated·May contain errors
Welcome to E-Journal Club episode 13.
I'm Ananth Kumar Kamala. What if we could take a
blood sample from a pregnant individual,
create a digital representation of her immune cells,
and explore how different drugs might change
those cells before testing a treatment in a
clinical trial? That is the idea behind today's paper,
Single-Cell Level Digital
Twins for Preterm Birth Prevention Strategies,
published in Science Advances.
The researchers developed a framework called SIMCO,
Simulated Immunome Modeling of Clinical Outcomes.
Their goal was to connect drug-induced changes
in immune cell signaling with predictions of when labor might begin.
Let's look at how they did this, what they found,
and where the science still needs to go.
During pregnancy, the maternal immune system changes continuously.
These changes help maintain pregnancy
while preserving the ability to respond to infection.
As labor approaches, immune signaling shifts again.
The important point is that this is not a simple
switch between immunity being on or off.
Different cells and signaling pathways follow different patterns.
The researchers asked whether drugs could shift
these patterns toward an immune state associated
with an earlier stage of pregnancy
and whether a computational model would interpret that shift as
a longer time until labor.
To investigate this, they first needed an experimental map of how
immune cells respond to drugs.
They collected blood from 11 pregnant volunteers between
approximately 20 and 27 weeks of gestation,
together with eight non-pregnant volunteers.
They then exposed the blood samples to candidate compounds
and different immune stimuli.
Using mass cytometry, they measured signaling
and activation across 28 immune cell populations.
Their atlas contained approximately 218 million
leukocytes across 80 perturbation scenarios.
That is an enormous number of cells,
but it is important to remember that those cells
came from 19 individuals.
Cellular detail and patient diversity are different strengths.
The screen included eight individual compounds and one combination.
There is an important pharmacological detail here.
For aspirin, the researchers
tested salicylic acid.
For folic acid, they tested 5-methyltetrahydrofolate,
abbreviated as THF in the paper.
So, when discussing the results, we need to distinguish
the compound tested in the blood assay from the medicine
or supplement administered orally.
How did these experiments become digital twins?
SIMCO worked in three main steps.
First, a computational method called CellOT
learned how immune cell measurements
changed after stimulation or drug exposure.
Second, the researchers applied
those learned transformations to baseline immune profiles
from a separate pregnancy cohort.
This created simulated versions of each sample under untreated
and treated conditions.
Third, another modeling method called STIBL linked the simulated immune
features to time until labor.
The researchers could then compare the predicted time to labor
for a treated digital twin with the prediction
for its untreated counterpart.
Here, a digital twin is a computational representation
of measured immune cell behavior.
It is not a complete virtual pregnancy.
The independent clinical cohort included 63 individuals
with only 5 experiencing spontaneous
preterm labor.
The model's predictions of time to labor correlated
with observed timing with a correlation
coefficient of 0.64.
However, the root mean squared error was approximately 24 days.
That tells us the model captures a meaningful relationship,
but its predictions are not
precise enough to treat as an individual delivery countdown.
What happened when the researchers simulated drug treatment?
5-methyltetrahydrofolate, maprotiline,
and the salicylic acid-lansoprazole combination
produced the strongest model shifts toward a longer gestation.
The authors particularly highlighted
the folate-related compound and the aspirin-related
combination as candidates for further investigation.
They also found variation between individuals.
A treatment predicted to help one digital twin did
not necessarily help another.
This is one of the most interesting aspects of the study.
It moves beyond asking whether a compound has
an average anti-inflammatory effect
and begins asking whose immune profile might respond favorably.
But the numerical results require careful interpretation.
For example, the reported effects of approximately 25 days
for THF and 17 days for the combination
were aggregated proxy effects from the model.
They were not additional days of pregnancy observed after treatment.
The study did not establish
that these interventions prevent preterm birth.
What biological signals drove the predictions?
Two important patterns involve naive CD,
four positive T-cells and granulocytes.
THF and the salicylic acid, lansoprazole combination reduced
STAT3-related signaling in naive CD4-positive T-cells
and increased S6-related signaling in granulocytes.
Within this model, these changes shifted selected immune features
toward patterns associated with an earlier stage of pregnancy.
This gives us mechanistic hypotheses to test.
It does not yet establish that changing
these pathways will cause labor to occur later.
The combination also deserves a careful description.
Most measured interactions between salicylic acid
and lansoprazole were additive.
Only a small proportion were classified as synergistic.
Therefore, we should avoid presenting
this as a broadly proven synergistic therapy.
From a reproductive pharmacology perspective,
I see another important question.
Do changes in circulating immune cells tell us what
is happening at the fetal-maternal interface?
The placenta, decidua,
and fetal membranes have their own cellular environments,
inflammatory responses, and barriers to drug exposure.
This study examined peripheral blood.
It did not directly measure how these
How these candidate treatments affected
fetal membrane cells or placental tissue?
It also did not model the full sequence of absorption,
metabolism, distribution and
elimination during pregnancy.
An assay concentration can help establish biological plausibility,
but it cannot tell us whether repeated dosing will maintain
the required unbound drug concentration
at the relevant tissue. This is where pregnancy pharmacokinetics,
tissue models and organ-on-chip
chip systems could complement immune digital twins.
A future workflow could begin with computational prioritization,
then test the predicted responses in fetal membrane
and placental models,
evaluate achievable exposure,
and finally move toward prospective clinical studies.
That is a future direction, not something demonstrated in this paper.
What should we take away?
Simcoe provides a proof-of-concept
for connecting single-cell drug responses with predictions
of a clinically important pregnancy outcome.
Its strengths are the detailed experimental atlas,
the ability to simulate heterogeneous
cell responses, and the attempt to explain
which immune pathways influence its predictions.
Its limitations include the small number of donors,
very few preterm cases, a restricted
gestational window for the drug response experiments,
and the absence of prospective validation
of treatment benefit.
Most fundamentally, changing a marker associated with labor
does not automatically mean that labor itself will change.
The exciting opportunity is to use these models
to make better choices about what to test next.
Can digital twins help prevent preterm birth?
This paper takes an important step toward answering that question,
but experimental and clinical validation must follow.
Please like, share and subscribe my channel.
Thank you for joining eJournal Club episode 13.
