Technology

Computing for biological function.

A signaling model advances molecular states. A mechanics model predicts motion under load. We develop computing and optimization methods around the equations and evidence each problem requires.

Connecting the models

What one solver needs from another.

Consider a proposed ionic actuator. A transport calculation might produce a concentration field. A mechanical model would need a tested relationship between that field and material strain to predict displacement.

Proposed solver workflow Conceptual diagram
  1. Define the experiment

    Specify geometry, boundaries, stimulus, output units and the reference observation.

  2. Exchange state

    Identify which solver owns each quantity, its coordinate system and when it is updated.

  3. Check the result

    Record solver versions, input provenance, conservation errors and agreement with the reference.

The lab runs specialized research components today. A shared experiment across independently qualified chemical, flow and mechanical solvers is a development goal.

Research capabilities

Where the technology stands.

SIML Compute

Demonstrated

Fit a scientific model into available memory and measure the cost of advancing it.

Current work
The lab has compared CUDA implementations of cellular signaling and binary lattice updates against tuned baselines. The two studies count different kinds of work.
Connection
A model supplies its state, update rules and correctness checks. The compute experiment varies the representation or execution while preserving that workload.
Limits
The reported gains apply to specified populations on one GPU. Transfers, setup, observation frequency and scheduling can change application performance.
Next test
Repeat a matched comparison on a partner workload and another GPU with a declared accuracy tolerance.

SIML Biochem

Implemented

Describe how a chemical state changes through reactions and transport.

Current work
Research solvers track chemical amounts through reactions and transport and calculate electrostatic fields. Analytical and conservation checks cover individual components. The separately published cellular experiment benchmarks one signaling model.
Connection
A coupling experiment would pass quantities such as concentration or flux to another model, with units, time intervals and conservation rules declared at the boundary.
Limits
Component tests leave the accuracy of a fully coupled tissue trajectory open. The published signaling benchmark has its own model, hardware and time horizon.
Next test
Choose an observable time course and test its prediction under conditions held out from calibration.

SIML FEP / Optimization

Demonstrated

Use feedback from a model to decide which action or candidate to evaluate.

Current work
A published planning experiment tested a surprise-based gate on candidate movements in a single-goal reaching task, comparing the same planner with the gate enabled and disabled.
Connection
We are investigating whether an objective that accounts for uncertainty could help select informative experiments. Each new objective needs a comparison with a conventional search strategy.
Limits
Success in a reaching task leaves open the performance of a materials search. Chemistry, fluid flow and mechanics each require an appropriate numerical solver.
Next test
Compare candidate-selection policies on held-out targets with equal evaluation budgets.

SIML Biological Models

Implemented

Choose a measurable behavior that can serve as a reference for an engineered alternative.

Current work
Separate research simulations represent movement under load, cellular signaling, visual control, deformable tissue and information propagation. Their validation targets and solvers differ.
Connection
Muscle provides a mechanical reference, vision a sensing and control problem, and vascular tissue a pressure–deformation relationship.
Limits
An anatomical rendering establishes geometry. Predicting a physiological response also requires measurements under stated loading or stimulus conditions.
Next test
Test a reference model against an independent measurement before using it to set equivalence targets.

SIML Discovery

Investigational

Compare candidate explanations or designs against an explicit functional target.

Current work
A virtual actuator study couples a mechanical simulation with candidate search. Separately, the causal-geometry study compared predictors of communication delays in simulated recurrent networks and reported unsuccessful tests alongside held-out results.
Connection
Candidate exploration would combine an experiment definition, solver outputs, comparison criteria and a record of how each candidate was selected.
Limits
Held-out prediction in simulated networks leaves candidate generation, material ranking and autonomous physical experimentation as separate research questions.
Next test
Run a prospective comparison of candidate rankings against an independent reference and preserve rejected candidates.

SIML Materials

Investigational

Investigate whether a different material or mechanism can deliver selected biological functions.

Current work
The materials workbench runs generic molecular and continuum reference models. An axial actuator model computes shortening and recovery from a prescribed response law under mechanical loading.
Connection
Ionic transport, electrochemical response and mechanics would have to be coupled for a candidate polymer to predict motion under load.
Limits
The low-voltage actuator is a research question. A physical performance claim requires characterized samples, controls and repeatable measurements.
Next test
Define the test load, drive constraints and accepted force and displacement errors before screening materials.

Working with the lab

Bring a workload and a baseline.

Set the memory budget, numerical tolerance or observation interval that matters to your team. We can discuss what a fair technical evaluation would require.