Stromalytix / Biofab Box
I’m building an automated loop for biotech R&D and biomaterials teams: research the literature, propose an experiment, simulate tissue behavior, then coordinate lab robots and compare real measurements with the prediction. Each cycle should improve the next design.
Liver is the first demonstration. The broader goal is a workflow adaptable to different tissues and biomaterials. Biofab Box is the planned hardware component, starting with a donated printer.
The software and models are demonstrated separately today. Connected robot control, physical validation and the complete automated loop are the next build.
Animal studies, simulations and 3D cultures each answer useful questions. Their limitations make reproducibility and validation central to the workflow.
Animal studies · translation
Differences in metabolism, target biology and immune response can limit how an animal result predicts a human response. Animal studies can identify some hazards; predictive value depends on the endpoint and context.
The need is for qualified human-relevant evidence, not an assumption that any one replacement method is sufficient. FDA’s 2026 progress report ↗
In silico · credibility
A computational result depends on its cell- and material-specific parameters, assumptions and boundary conditions. Calibration, uncertainty assessment and comparison with suitable measurements are needed for the intended use. A simulated shape or oxygen field alone does not establish biological function.
Our CC3D study is uncalibrated; the oxygen model is separate. FDA model-credibility framework (devices) ↗ · Draft NAM validation principles (drugs) ↗
3D culture · materials & reproducibility
Some widely used animal-derived matrices have incompletely defined compositions and lot variation. Matrix chemistry, adhesion and stiffness can affect cell behavior. Cell source and handling also need controls; moving to 3D does not by itself standardize the test system.
Defined synthetic matrices address some of these issues. The requirement is to characterize the actual formulation and lot. Matrigel proteomics ↗ · Three-batch proteomics study ↗ · Defined-matrix research ↗
3D culture · dimensional stability
A target dimension is not a measured dimension. Depending on the material and conditions, constructs can swell, shrink or remodel after fabrication. Geometry and mechanical stability should be checked over the assay period so the model uses the test system that actually exists.
One gelatin–alginate scaffold study observed concentration-dependent shrinkage within hours, followed by stability; this is an example, not a claim about every hydrogel. Primary scaffold study ↗
Implementing a biological protocol on robots also requires programming, motion and dispensing calibration, material qualification, imaging and data integration. A biology team may have to recruit or develop expertise across all of these disciplines.
Stromalytix is intended to connect that work so the scientist can specify the experiment and inspect its evidence. It does not remove the need for engineering or scientific qualification. Biological lab-automation study (2023 preprint) ↗
Liver injury is a consequential drug-development failure mode. It gives us a focused first use case for connecting material selection, oxygen transport, fabrication and function. The intended workflow extends across tissues and biomaterials.
The first physical milestone is repeatable printing and measurements that challenge the models. Drug-safety performance requires additional biological validation. FDA DILI context ↗
The parameter library, actual StarNet agent work, executed models, hardware designs and the proposed path to drug-safety evidence.
3:30 · Silent walkthrough with on-screen explanations. The software is demonstrated; printing and biological validation are next.
Choose a chapter, then press Play. Use fullscreen for small text, or open the written walkthrough below. Download the video.
The public parameter library lets a scientist inspect the number, unit, material, cell type, source label and confidence before using an input.
A literature-linked entry can include a conversion or assumption. Model estimates are not measured constants. The Biofab review checks these distinctions before accepting simulation inputs.
Actual model-backed activity recorded in an isolated native StarNet workspace. Two Biofab review tasks, public inputs and saved draft reports.
The parameter reviewer reads six records and checks source labels, unit conversions and calibration gaps. The assay reviewer maps proposed modules and external tests to tissue quality, liver function and drug-safety follow-up.
Both agents saved draft files. The parameter run completed; the assay run saved its file, then was canceled at a subsequent deliverable-note approval. These were sequential tasks, not an autonomous lab run.
Review caught an unsupported claim about the absence of published measurements and an incorrect claim that imaging was already integrated. Both drafts require corrections before scientific acceptance.
Execution is observable. Scientific qualification remains a separate gate. The corrected assay map and model limitations on this page describe the current scope.
Executed 3D CompuCell3D sensitivity study. A local mechanism experiment for the liver-tissue design context.
At 600 Monte Carlo steps, matrix contact was 405–411 lattice faces for J = 2 and 219–221 for J = 24. Everything else was held fixed. These are ranges across three seeded runs, not confidence intervals.
Cell count, connectedness, volume bounds, matrix stability and exact same-seed repeatability passed numerical checks.
The cells and rigid interface are an idealization. No parameter mapping to HepG2/C3A, a GelMA formulation, micrometres, hours or stiffness has been established. This is a complementary sensitivity study, not the same geometry as the oxygen model and not a coupled simulation.
No viability, liver function, drug response or formulation recommendation follows from this demonstration.
How does thickness change the oxygen available inside a printed construct?
Switches between previously calculated cases. This page does not rerun the solver.
Nominal minimum oxygen in the gel
This separate reference calculation uses HepG2/C3A assumptions and a GelMA reference material. It models steady oxygen transport and cellular uptake in a static well.
The range reflects eight combinations of uptake, diffusivity and surface-oxygen assumptions. It is not a confidence interval. The next decision is which measurements will narrow those assumptions.
Numerical checks passed for the saved calculation. Physical printing, oxygen measurements, cell viability and liver-function testing have not been performed.
Saved calculation: 67f4272f0fb6bbf3
The model is a 2D axisymmetric finite-volume calculation with Michaelis–Menten oxygen uptake, separate gel and medium diffusivities, fixed surface oxygen, and no-flux walls and floor. It assumes uniformly distributed cells; it does not model growth, death, perfusion or drug response.
The software workflow in the video extracts seven source-linked inputs from a published primary-human-hepatocyte study. This oxygen calculation is a separate HepG2/C3A reference, not a validated prediction of that paper’s experiment.
Recovered design iterations and module concepts. These images show proposed hardware, not completed instruments.
I assembled a wooden Pitsco T-Bot hydraulic arm. It gives me a physical starting point for exploring manipulation and transfers alongside the printer conversion.
Motorization, sensing, repeatability, contamination control and safe integration remain engineering work. The kit’s manual hydraulic motion is not autonomous laboratory handling.
The actual donor is a Creality CR-10. Temperature-controlled dispensing, automated registration, perfusion, mechanical testing and imaging integration are design goals. The initial laboratory workflow includes manual transfers.
Planned readouts for the liver use case. Each requires a qualified method, controls and the right instrument; the Biofab Box does not itself perform every assay.
Imaging + suitable optics / stains
Brightfield tracks dimensions and gross morphology. Fluorescence can map membrane integrity and live/dead staining through the construct. This separates printing defects or baseline cell loss from a candidate drug effect.
Tissue QC and a cytotoxicity screen; viable cells alone do not establish liver function.
Sampling + external plate reader / reagents
ATP reports cellular energy associated with viable cells; LDH release reports loss of membrane integrity. Agreement across distinct endpoints helps assess concentration-dependent cytotoxicity and flags compounds needing closer investigation.
Assay interference, transport limits and untreated controls must be checked. Viability methods ↗ · Cytotoxicity methods ↗
Culture / sampling + external assays
Albumin and urea measure aspects of hepatic function; changes after exposure can reveal dysfunction even when cells remain alive. An injury marker such as ALT adds a complementary damage signal.
Use cell-source-appropriate benchmarks. Published human Liver-Chip study ↗
Qualified hepatic model + assay / analytical access
CYP assays assess drug-metabolizing activity. Appropriate inhibition or induction studies inform interaction-risk follow-up. Analytical measurements such as LC–MS can check actual compound exposure and metabolites when required.
A CYP readout alone is not a complete interaction assessment. FDA / ICH M12 ↗
Calibrated mechanics + oxygen measurements
Compression and relaxation characterize the gel; direct oxygen measurements challenge the transport model. These controls help distinguish matrix or culture artifacts from biological responses to a compound.
Mechanical stiffness is a material property, not a liver-toxicity assay.
Decision supported after validation
Combine reproducible concentration–response data, functional readouts and exposure context to prioritize compounds or request confirmatory studies. Establish performance against reference compounds, relevant human cells and independent batches first.
The current HepG2/C3A engineering model has not established human drug-safety predictivity. No safe/unsafe drug conclusion is demonstrated here.
SystemLens + StarNet OS lab-integration concept. This local mockup illustrates task handoffs and a digital-twin layout; it is not connected to the live StarNet service or laboratory equipment.
Biofab BoxPlanned syringe conversionCulture stationLab access still requiredMeasurement stationOxygen, geometry and assaysStep 1 of 6
Inspect the 0.2 mm case and its assumptions. Proposed dispensing target: 0.402 µL. No machine command has been sent.
The intended process record links a sample ID, protocol version, equipment calibration, source artifacts, operator handoffs and eventual raw measurements. This interface demonstrates that planned control flow. Conversion, calibration, validated operating settings and qualified lab access are prerequisites for a physical run. Completing the walkthrough creates no experimental results.
A proposed verification study exported alongside the geometry and predictions.
The proposed study has separate 24-hour and day-7 endpoints. Batch is the independent unit. Lab-specific qualification and operating procedures must be completed before execution.
Agreement would support the oxygen model within the tested system. It would not establish drug-safety accuracy or biological function.
The scale of drug-development spending makes earlier, trustworthy evidence valuable. These figures are context, not a validated addressable market.
of treatment candidates entering clinical trials fail, according to NIH NCATS. This includes efficacy and other causes, as well as safety.
NIH NCATS · strategic plan ↗spent on pre-human / preclinical R&D by PhRMA member companies in 2024, within $104.3B of total R&D.
PhRMA 2025 survey · tables 1 & 3 ↗FDA draft guidance sets out validation principles for human-relevant alternatives, including tissue models and computational methods.
FDA · March 18, 2026 ↗Clinical attrition has multiple causes. The spending figure is an industry budget, not Stromalytix’s TAM; FDA guidance is not approval of this prototype.
The business hypothesis is a repeat-use platform for pharma, biotech and contract research labs: software, instrument support and consumables.
Start with a defined liver-tissue experiment and a laboratory partner. Measure repeatability, operator time, cost per usable construct and agreement with reference assays. Expand to more workflows and tissue types as each is qualified.
Illustrative recurring-revenue scenario
100 sites → $10M/yearAt an assumed $100,000 per site per year in recurring software, support and consumables, 1,000 sites would represent $100M/year. One-time hardware sales are excluded.
These are arithmetic scenarios. Pricing, reachable site count, retention, margins and willingness to pay remain unvalidated; they are not revenue, customer traction, a forecast or a market-size estimate.
The next milestone is a calibrated printer and measurements that can challenge the saved predictions.
Biomaterials developers need to connect a formulation and its operating conditions to the experiment it is intended to support. Stromalytix is being built to make those inputs, assumptions and evidence inspectable.
Biotech R&D teams need reproducible experiments and a clear record of what changed. The intended workflow connects design decisions to fabrication and measurement, then uses the results to improve the next design.
The public repository may lag this prototype. The downloadable calculation and protocol above document the run shown here.
