From Sequence to Function:
Engineering Molecular Devices for Monitoring Living Systems

September 2026 MAE Seminar

Jing Pan
Assistant Professor
Mechanical and Aerospace Engineering

beat 0 - opening: what the talk is about

Thank you Hugh for this nice introduction. And thanks everyone for attending today’s seminar.

I want to share my lab’s research over the past several years: building useful devices at a really small scale, about the size of molecules.

As mechanical engineers we are used to building things we can see and hold. The systems we work with are millions of times smaller. The two research questions I always ask myself are how we can rationally build machines that small, and what they are good for. You may remember all the pop culture images of submarines going through the bloodstream doing amazing things.

I started my lab in 2020, right around when COVID happened, and everyone was thinking about better ways to measure biological events in living systems.

So my lab started looking into the engineering science of building molecular-scale devices, and also how they can be useful as biosensors for monitoring living systems.

And that is the story I am going to share today. Let me start with an example of a molecular machine.

Talk Overview

1. Introduction: Biomolecular Nanotechnologymolecular machines, applications, engineering science, challenges
2. Molecular EngineeringAI-assisted design, high-throughput screening — sequence to function mapping
3. Continuous Biosensingbiosensing trilemma, photothermal recycling, continuous microfluidic flow assay
4. Microphysiological System Integrationmonitoring living cells on chip, systems biology of immune response

Molecular Machine: Myosin V Example

step36 nm, hand over hand

detachment force~2.5 pN

rate~5 steps/s

fuel1 ATP per step

run length~1 µm

Transport

High-speed AFM

Hearing

Muscle

Cell mechanics

Myosin slide:

This is myosin V. It carries vesicles along actin filaments.

It walks hand over hand, thirty-six nanometers per step, one ATP per step, about five steps a second. It detaches at around two and a half piconewtons - it lets go rather than stalling.

This is high-speed atomic force microscopy, from Toshio Ando’s group. Each frame is about eighty milliseconds. Nothing here is labelled or reconstructed. This is the molecule itself, walking.

Fantastic Molecular Machines

beat 2 - fantastic molecular machines

Machines like this run most of what happens in a cell.

A replicative polymerase copies DNA at close to a thousand bases a second, and makes an error about once in a hundred million. ATP synthase turns at over a hundred revolutions a second and produces most of the ATP you use today. The bacterial flagellar motor spins at a couple of hundred hertz, and reverses direction in under a millisecond.

Fantastic Molecular Machines and Where to Find Them

Fantastic Molecular Machines and How to Build ThemWhere to Find Them

• Materials and mechanics
• Control and dynamics
• Power and energetics
• Fabrication and manufacturing

De novo
design

Applications: Now and Future

NOW

FUTURE

Chemicals & Energy

Detergents, plastics, fuel

Defense

Detection and countermeasures

SMART-RBC

PROTEAN

Agriculture

Natural insecticide corn

Biomedicine

Vaccines, cell therapy, CGM

beat 3 - all four at once (the one-slide version; use this OR the four detail slides, not both)

Some of these are already engineered, and already in use, and it is worth seeing all of it at once.

Top row is now. Engineered enzymes run cold-water detergent and a large share of industrial catalysis, and Cargill’s lactic acid becomes the plastic cup you drank from this morning. In defense, what is actually fielded is measurement - a biothreat panel that is engineered primers and a polymerase, and an anthrax stockpile. In agriculture, engineered Bt and herbicide-tolerance proteins are on more than ninety percent of US corn, cotton and soy - that is not a pilot, that is the landscape you fly over. And in medicine, mRNA vaccines, cell therapy, and a continuous glucose monitor that is an enzyme on a wire under the skin.

Bottom row is what is being paid for next. Biorefineries and enzymatic plastic recycling. Manufacturing a protein at the point of need in twenty-four hours, and releasing engineered microbes as environmental sensors. Cereals that fix their own nitrogen. And continuous measurement of something other than glucose - hormones, drug levels, inflammation.

Four separate fields, four separate funding agencies, and all of them are buying the same missing capability: specify what a molecule will do before you make it.

Gap and Research Needs

NOW(natural)

FUTURE(de novo)

?

1.Engineering science

2.Measurement tools

3.Datasets and standards

4.Design and prototyping capabilities

TODO - the middle column is empty on purpose. Left is what biology already made and we found. Right is what can now be written from scratch. The gap between them is the argument of the rest of the talk, and it goes in the middle box once we settle what it says.

The Convergence of Energy Scales

Electrostatic shell

Bending, 20:1 rod

Fracture, 20:1 rod

Thermal

Electron in a box

beat 6 - found, not designed, and why

That one sensor was found, not designed.

Glucose oxidase came from a mold. Most binders used in diagnostics come from immunizing an animal, or from screening a library of billions of random sequences and keeping whatever binds. Directed evolution works. It takes years per molecule, and what you get is whatever the search happened to find.

Rubisco is the clearest case. It is the most abundant enzyme on Earth, it fixes essentially all the carbon in the biosphere, and decades of work have not produced a better version.

The reason is worth a minute.

Consider a car. A fuel tank stores chemical energy. An engine converts it to motion. A frame and drivetrain hold the geometry and direct where that motion goes. Three components, three functions, each confined to its own place and its own moment.

Now take the same machine down to a hundred nanometers. Thermal energy is comparable to the work done in one cycle, so the frame deforms continuously. A structure that size moves at a couple of meters per second from heat alone. And the frame is also the fuel tank and the engine. The same atoms bind ATP, hydrolyze it, and carry the mechanical load.

At this scale mechanics, chemistry and thermodynamics are the same degrees of freedom. Functions cannot be separated in space.

Biomolecular Nanotechnology

Parts

design, screen and abstract functions into standard parts

Nuclease

Ligase

Polymerase

MEDAKNIKKGPAPFYPLEDGTAGEQLHKAMKRYALVPGTIAFTDAHIEVNITYAEYFEMSVRLAEAMKRYGLNTNHRIVVCSENSLQFFMPVLGALFIGVAVAPANDIYNERELLNSMNISQPTVVFVSKKGLQKILNVQKKLPIIQKIIIMDSKTDYQGFQSMYTFVTSHLPPGFNEYDFVPESFDRDKTIALIMNSSGSTGLPKGVALPHRTACVRFSHARDPIFGNQIIPDTAILSVVPFHHGFGMFTTLGYLICGFRVVL

Composition

design and build functional modules; test, learn and iterate

Integration

embed and couple with a non-sequence environment

Biosystem

Nanomaterial

Device

beat 8 - the method: biomolecular nanotechnology

They can be separated in the sequence.

A protein or a nucleic acid carries its functions in a string. Different stretches of that string do different things, in the same molecule. That is what biomolecular nanotechnology is built on, and it gives three levels of design. The useful question at each level is what you are actually specifying.

Parts. The variable is the sequence. One molecule, one function: bind this target, switch on that signal, hold this geometry.

Composition. The variable is how parts are coupled. Connectivity, spacing, order, feedback. This is where behavior in time gets specified.

Integration. The variable is the environment and how it couples back to the sequence. Surfaces, particles, flow, tissue.

Structural DNA nanotechnology and protein design built the first two levels. Synthetic biology also works at those two levels, with the cell executing the program. Here there is no cell. The molecule is the device.

beat 9 - the tools arrived, the measurement did not

None of the three levels is finished. What changed in the last few years is how much is available at all of them.

At the parts level you can cut, join and copy sequences to order. Structure prediction is accurate. Binders can be designed against targets that never had one. At the composition level you can fold hundreds of strands into a defined shape, run logic on them, and place proteins on a scaffold at nanometre spacing. Making the molecule is no longer the constraint - a sequence goes to a synthesis company and it arrives in a week.

What has not kept up is measurement. These methods predict structure, and binding strength at equilibrium. They do not predict rates. Function is still measured one sequence at a time, in whatever buffer the assay happens to use, so there is no dataset large enough to learn the sequence-to-function relationship from. That is true at the parts level and it is true at the composition level. Neither one is solved.

There is one exception, and it is the reason the first part of this talk exists. Sequencing has been inverted into a measurement instrument - a hundred million molecules read in a single run, each one linked to what it did. That is the shape the missing measurement has to take, at every level.

beat 10 - integration is the least solved

Integration is the least solved of the three. That is not the same as saying the other two are done.

The standard construction today is to take a molecule that someone found, attach it to a surface or a particle that someone else made, and measure what comes out.

At this scale the environment changes the molecule’s energy landscape. Tethering a binder to a surface shifts its effective concentration by orders of magnitude. Molecular crowding changes folding free energy by several kT, which is the same size as the function itself. Local heating changes the dissociation rate directly. The sequence is unchanged. The function is not.

So the performance of a molecular device is not a property of the molecule. It is a property of the molecule and its environment together, and only one of the two is written down. In the rest of mechanical engineering a solution means nothing until the boundary conditions are stated.

The strongest form of this coupling is when the environment supplies work to the molecule rather than perturbing it.

beat 11 - transition into part one

That is the framework. The rest of the talk follows it, in three parts.

First, the parts themselves - how we design them and how we screen them. AI-assisted design on one side, high-throughput screening on the other.

Second, putting those parts into a device and using them for continuous biosensing.

Third, putting that device against living tissue, and monitoring microphysiological systems.

About fifteen minutes each. Please stop me if something is unclear.

Molecular Engineering

AI-assisted design and high-throughput screening — mapping sequence to function

Continuous Biosensing

The biosensing trilemma, photothermal recycling, and continuous flow assays

MPS Monitoring

Microphysiological systems — monitoring living cells on chip