Form
MODEL:
Form
MODALITY:
3D MORPHODYNAMIC TRAJECTORIES
Seeing treatment response before it happens
Drug response doesn’t begin in gene expression or clinical outcomes - it begins in shape. FORM is our generative 3D framework that simulates how living cells reshape and reorganise under treatment, letting scientists observe the morphological consequences of a drug before that drug is ever given.
At Sentinal4D, we believe that understanding biology in motion starts with understanding how structure changes under perturbation. FORM gives us the first principled way to simulate these changes directly from real 3D single-cell microscopy, turning real biological data into foresight: a way to explore how cells are likely to respond under new drugs, doses, or genetic backgrounds.
FORM doesn’t classify, label, or predict in the statistical sense. It simulates treatment effects - faithfully, stochastically, and at population scale - using generative modelling grounded in real biology.
This is morphology as foresight.
This is morphological intelligence.
What FORM is
FORM is a two-stage generative system trained on large-scale 3D microscopy data:
A morphology encoder
A multi-channel VQGAN learns discrete prototypes of cytoplasmic and nuclear structure, ensuring that subcellular detail is preserved and biologically coherent.A perturbation-conditioned diffusion simulator
A latent diffusion model generates plausible post-treatment morphologies, capturing the full heterogeneity of how cells tend to respond under a given perturbation.
Together, these components allow FORM to simulate how 3D structure changes under drugs or genetic perturbations - at the level of individual cells and populations.
Why simulation matters for drug response
Biology is heterogeneous. Two patients with the same tumour may respond differently to treatment; even cells in the same dish diverge in their behaviour. FORM embraces this heterogeneity by simulating many plausible morphological outcomes for a single treatment, enabling scientists to:
See likely treatment responses before an experiment is run,
Explore how different genetic backgrounds may change a drug’s effect,
Identify structural features of resistance or sensitivity,
Test biological hypotheses virtually,
Accelerate early-stage decision-making by replacing uncertainty with evidence.
FORM does not claim to be deterministic. It doesn’t tell you what will happen - it helps you see what is likely, based on real 3D biological data.
What FORM can simulate
Unconditional simulations: replicating real perturbation landscapes
FORM synthesises realistic 3D morphologies for each drug treatment, matching:
the diversity of shapes seen in real experiments,
the fine-grained subcellular detail, and
the distributional structure of real perturbation responses.
Across FID, precision–recall, and coverage, FORM outperforms baseline generative models trained on the same data. These simulations act as virtual replicas of perturbation-specific cell populations.
Conditional simulations: “How would this cell respond to treatment?”
Given an untreated 3D cell, FORM simulates how that cell is likely to respond under a specific drug. These simulations recover known treatment responses.
Across thousands of simulations, descriptor distributions align with real perturbation data - meaning FORM captures not only the typical effect, but the spread of responses.
This is what makes FORM valuable for understanding heterogeneity.
Generalisation to unseen cancer subtypes
FORM can simulate how cells from one biological context respond to a treatment, even when that context was never part of the training data.
Simulated morphologies align with real responses from those new contexts, showing that FORM adapts treatment effects across diverse cellular backgrounds.
This capability is essential for exploring drug responses beyond the systems originally profiled.
Morphodynamics: simulating structural evolution through time
Real cells do not change shape smoothly - they shift abruptly, non-linearly, and stochastically. FORM captures this by reconstructing intermediate states across the diffusion process:
descriptor curves show non-linear structural change,
intermediate 3D volumes reveal evolving morphologies,
simulations more closely resemble real live-cell dynamics than interpolation baselines.
This enables virtual time-lapse studies of how structure evolves during treatment.
Simulating intracellular signalling from morphology
FORM can also simulate signalling activity directly from 3D cell structure, linking morphology to intracellular pathways in a generative way.
Simulated signalling readouts show strong correspondence with independent biochemical measurements, preserve the relative ordering of perturbation effects, and recover known pathway-specific responses.
This makes FORM a cross-phenotype simulation tool — one that moves from structure → signalling → potential treatment sensitivity.
How FORM supports Sentinal4D
FORM is a core component of the SentiCORE platform. It enables:
Virtual treatment exploration
Simulate how cells might respond to specific drugs before running an experiment.
Evidence-based decision support
Simulations reveal structural consequences of treatment that teams can inspect and verify.
Mechanistic insight
Simulated morphologies highlight the shape features linked to sensitivity, resistance, and signalling changes.
Context transfer
Simulate responses in new cell types or disease contexts.
Multi-phenotype integration
From shape to signalling, FORM connects multiple biological readouts through generative simulation.
FORM gives decision-makers clarity in a landscape defined by uncertainty.
What comes next
FORM establishes the ability to simulate treatment effects in 3D. Our next step extends this foresight into 4D - capturing real morphodynamics and treatment-induced trajectories across time.
This directly informs MorphoSense, our temporal framework for modelling dynamic shape change and interpreting 4D morphodynamic signatures.
Closing thought
Cellular change begins quietly; a shift in shape, a rearrangement of structure, a subtle response to treatment. FORM lets us see those changes ahead of time, simulating how cells are likely to behave under therapy, condition by condition, context by context.
If your work depends on understanding treatment effects before they unfold, FORM offers a generative, scientifically grounded way to explore that space. We’d be glad to discuss how.
