Autonomous Labs Create New Instrument Niches to Fill

When I give tours of our Nebula autonomous lab installation, one thing I like to note are instruments we have integrated that are workhorses for us yet still could be a bit better.  A running theme here is that most laboratory instruments were developed for walk-up use by individual scientists, with some automation integration capabilities grafted on to enable workcell use.  Or they were designed for workcells.  In either case, issues for autonomous operation arise such as being far too batch oriented, rather than asynchronous experimentation, as well as requiring skilled manual servicing.  Creating new instruments for an autonomous laboratory world is a great open innovation opportunity for academics, startups, and other mechanically-minded persons.  Today I'll talk about reagent filling.

The Status Quo

In a biochemistry experiment I've run on Nebula (yes, that is a story I need to tell soon!), two 384 well plates must be filled with reagents, one with reaction buffer and one with stop solution.  In this case, every well on a given plate has the same volume, and these volumes are much too large to use the Echo acoustic dispensing.  In another experiment, an engineering experiment to determine a correct pipetting setup (wait? Keith has successfully run two wetlab experiments??? Long-time readers deserve details!), I required a stairstep pattern on a plate, with the first column empty and each successive column with 5 microliters more than the previous column.  Those are reagent filling jobs.

One solution is manual filling with a multichannel pipetting.  That is tedious and for me absolutely nerve-wracking, even without the 16-channel dropping tips or having two tips engage in touching.  Plus the one plate required two passes due to the volume required and the maximum capacity of our 16-channel pipet, so twice the "did I hit every column once and only once" anxiety.

Another solution is a manual 96-well pipettor, in my case called a Liquidator.  Four stamps, using a special plate that allows accessing a 384 well plate by "quadrants" (not quadrants of the plate, but thinking of four wells as a set).  That's pretty easy and if you think of the four positions as NSEW easy to keep track of.  But does require presence in the lab.  

In both cases, the source is a disposable reservoir, a narrow V-shaped one for the 16-channel and a big rectangular one for the Liquidator.  Reservoirs are a bit tricky on an autonomous lab, with concerns of instability when cornering on the tracks.   With a reservoir, I could treat our Bravo or Lynx liquid handlers as reagent dispensers just like the liquidators.  

On Nebula, we have two reagent dispenser types plus platewashers which can be pressed into service.  The MultiFlo has options of syringe pump or peristaltic pump.  The Tempest uses microfluidic diaphragms for extremely precise reagent dispensing.

Both were designed for walkup or batch operation.  You, or an arm, places a plate on the system, press a button (physically or virtually), and the desired fill pattern is quickly executed.  But that requires prior entirely manual setup of the instrument to put the intake tubes into the correct reagent sources and later entirely manual work to clean the lines and remove the input vessel.

There's also the issue of the number of reagent channels available.  Each MultiFlo can be configured with four reagents and each Tempest with twelve.  It's not hard to cook up experiments that require far more custom reagent fills than my current two; a different biochemistry experiment I've been designing needs four.  True autonomy means not having to worry if I'm being piggish with reagent dispensing slots, many of which may be allocated to staples such as common buffers or frequently used mastermixes or all the favorite flavors of functionalized magnetic beads.  If I want to run four reagents and ten of my colleagues also want to each have 4-6 custom reagents, we soon have a logjam.  True autonomy also means not requiring someone to manually switch-over a dispenser from one reagent to another; that should be autonomous as well.  

There are other reagent dispensers on the market, and I think they all follow the same pattern - manual setup and takedown.  One startup to watch is OpenShelf, which may be trying to break out of these patterns.  But what are new patterns?  Ones that might fit any autonomous laboratory scheme.  I'm of course paid to be a chauvinist for our particular implementation, but I must acknowledge the existence of others (even if I won't give them free advertisement :-) 

So here are some sketches intended to be the stone of Stone Soup, seeding the idea pool but don't let them limit you!  And they certainly can be recombined in various ways.

Robotic Operation of State-of-the-Art

One solution would be more sophisticated robots to work the existing instrument choices.  There's all sorts of small tubes and little fittings and more to work, which can't be an easy robotics task - but what robotics engineer wants an easy robotics task?  Some MultiFlo units in non-working condition are going for as little as $495 on eBay; I found a probably derelict Tempest for $295.  Even demonstrating working with a dead instrument would be a nice proof-of-concept for a reagent filler robo-jockey.

Sealed Reservoirs

What if we had fully sealed reservoirs that could flawlessly travel around the tracks or any other motion system, but then could be used by a compatible reagent filling instrument?  A bit like lunchbox juiceboxes: they travel well but can be sipped from when needed.  

Reagent Depot

What if instead of the juiceboxes traveling around the tracks, there was just a large storehouse and filling station.  Ideally new juiceboxes could be added, and empty or unneeded ones removed, without halting the instrument.  Perhaps it looks like an Automat, with little cubbies for each juicebox.  Or perhaps it is more like a two-way vending machine, able to accept or emit juiceboxes.

 

Further Considerations

In designing a solution, don't get trapped thinking it will be perfect.  None of these are, and so your new device needn't be either.  Having an option to refrigerate the reagents would be super valuable, but neither MultiFlo nor Tempest has that built in.  Perhaps for this version don't worry about reagent dead volumes (Tempest has some amazing capabilities there), but keep them in mind.  Similarly, don't get too hung up on the minimum reagent dispense or precision; those are very important properties but can probably be honed later.  Equipment cost or disposable costs should similarly be punted down the road.  Well, don't make things out of unobtanium but otherwise cost should be secondary.  Just get some new ideas out there!

Of course, having a solid and modern software API is critical.  Unique protocols should be able to be pushed up to the instrument via API and should be inspectable on the instrument, with cryptographic signatures and versioning.

Ladies and gentlemen, start your tinkering!