Liquid Class Optimization

Liquid class optimization can be time-consuming and annoying, especially if you have many different demanding liquids (high variance of vapour pressures, volumes, viscosities,…).

I wondered what’s your tips and tricks for liquid class optimization to ensure good precision and accuracy and high robustness but also avoid drop formation or even dripping while increasing pipetting speed?

How do you usually proceeed to develop a new liquid class from scratch?

Which ones are the key parameters of liquid classes?

Can liquid classes be generated automatically or can liquid class parameters be edited by a script or software to automatically test a bunch of parameters over night (e.g. with automated gravimetric detection) to optimize precision and accuracy of the pipetting step?

Has anyone ever used a DoE approach for that?

I’m curious about your experiences…


I’m also curious about what people do. I’m sure a DoE approach would be possible. You could use three responses using gravimetry such as: (1) mass of the droplets after holding the pipet still above some plate that is weighted and (2) pipetting accuracy and (3) pipetting precision. Maybe depending on the fluid viscosity and other liquid characteristics you can pick individual factor ranges. You would need a separate model that gives you a good guess of what ranges are needed. A such protocol for automatically generating liquid classes would optimally just require a few user query parameters such as “estimated viscosity”, “estimated volatility”, “concentration”, etc.
What I dislike about Venus 5 though is that you cannot edit the liquid class parameters dynamically. You can only work with pre-generated liquid classes that can be dynamically mapped to your pipetting steps though. They should definitely implement that.


You can modify liquid classes at run-time. It is not obvious but the liquid class parameters are in a database so running SQL scripts to modify those parameters works.

Hamilton did have a method they used with their integrated Balance to do what you are describing about walk-away liquid class optimization. This was 5-6 years ago and I don’t know how good it was but it would just repeat pipetting steps, measure, and then adjust the liquid class and repeat.

Most of our clients use ARTEL or hand pipette qualification of their liquid transfers.


Has anyone ever done gravimetric calibration with a digital scale? This seems a lot cheaper than Artel dye based stuff and could be totally closed loop


Yes Hamilton service engineers usually have a Mettler Toledo balance that has data accessible through Venus. You can reach out to your Hamilton rep about borrowing it for liquid class testing or buying one from Hamilton.

1 Like

This is true, you can actually modify liquid class parameters at run time on Hamilton which is clutch if you hate your data.

On a Tecan (Fluent) I believe you can also adjust accuracy dynamically but it’s not widely utilized. I can’t speak for other instrument types.

1 Like

Thanks for the helpful answers. I personally do not have that much experience with SQL, is there anything to keep in mind when trying to modify the liquid class?

We currently do use a balance to measure the analytical performance of the steps, but to unleash the full power of this approach we definitely had to do it in a fully automated loop. Manually varying the pipetting parameters is a mess. I think to automate it could also be helpful for generating a robust data set through repetition and variation to be able to apply statistical methods for data evaluation.

In general I am happy with the gravimetric approach: you receive a direct response, also resolution of a proper balance is good enough. Most balances can be connected via USB or RS232. But they are quite expensive, the said Mettler Toledo is >10k€.


We use Bayesian optimization in our lab quite often for these types of problems, and I’ve seen other labs use it for optimizing HPLC methods. So, I think BO would work for optimizing liquid classes.

I’ve got a package called Summit that makes BO quite easy to use, especially when you have multiple objectives. Would be happy to create a quick script in python if someone wanted it.


In my experience, 90% of the time the default liquid classes (in Venus) are pretty darn good. As long as you’ve got the correct liquid type selected.

If you find that it’s off, there are a few way to check your pipetting. Gravimetrically (like what’s already been mentioned), colorimetric testing, you could even just check it with your manual pipettes or eye test. Really depends on the level of precision and accuracy that you need.


At my work, we (Festo) benchmark everything using calibrated scales. It is the gold standard. The drawbacks are you can’t test multi dispenses, and proper scales are ~$12k with another 1-2k per year in calibration and maintenance costs. Artel has the big plus over gravimetric calibration in that they can of read/test entire plates. There are other ways too, but Artel has managed to become the standard, IMO.


Artel is pretty expensive. They’ve also lobbied extensively to set the standards I believe.

They’re also raising their prices…

1 Like

Is there any reason that food dye plus a plate reader wouldnt work just like an Artel? As long as the formulation is very consistent for a dye that seems like it would provide very similar results

1 Like

There’s definitely room for an alternative and the answer is no. There are known dyes that can work comparably to what they provide. The problem is that for certain lab environments they meet regulatory requirements.

Prior to ARTEL adoption, you could kind of get away with Tetrazine and a Spectramax.

Synthace have done DoE liquid class optimisation with their software. Really smart way to do it if you have the capability

These guys might be able to provide an alternative to Artel if you need somthing more than food coloring:

1 Like

Hi all,

As others have alluded to on the thread, while Hamilton provides tools to measure liquid transfers, we do not offer a fully automated liquid class development solution. I agree that this would be beneficial for our internal support teams and our customers.

See below for solutions Hamilton currently provides for additional liquid class development and verification:

  • Liquid Verification Kit (LVK) - gravimetric approach using the same balance as Hamilton’s field service. Includes deck hardware and a GUI with VENUS method to develop and verify liquid transfers.
  • Hamilton Artel Verification Kit - this is in progress. Hamilton and Artel are collaborating on a standardized method and optimized liquid classes to use the Artel MVS to assess performance of the Hamilton STAR or VANTAGE liquid handlers. We’re in the middle of compiling performance data across multiple sites before formal launch.

In regards to general guidance on liquid handling on our platforms, please see this link for the Liquid Handling Reference Guide. The guide provides an overview of common liquid properties, how to account for such properties from a hardware and software standpoint, how to monitor transfers for exception, and finally, how to verify liquid transfers. That last section goes into more detail about dye based and gravimetric approaches including the pros/cons of each. As an aside, after working and learning more about the Artel MVS during our recent collaboration, this section could likely use some updates/corrections.

I hope you find this information to be helpful!

Thank you!



Always love to hear from Hamilton folks themselves! Thanks so much for sharing this. I’d also massively appreciate any encouragement to other Hamilton employees to share their expertise here as it is very highly valued by the community.


Thank you for the valuable input. I appreciate your contribution very much. Does Hamilton also follow the approach of optimizing liquid classes by DoE strategy?
What’s your opinion on dynamic editing of liquid classes via SQL access to automatically (gravimetrically) test variations of the liquid class parameters, as described earlier here? Do there exist appropriate Hamilton libraries to do that?


Hi Max,

When we develop liquid classes in the field, we’re not following a DoE strategy - typically, we’re selecting a default liquid class from a liquid type that is similar and making incremental adjustments to the parameters to achieve good precision and then adjusting the correction for trueness. Adjustment to which parameters depends on visual observations and measurements with a balance - for example, if there are hanging droplets, then adjust the air transport volume. Often, we may only be interested in a particular volume for a specific application so we may not spend the time checking and optimizing at other volumes for the selected tip type.

For our internal LH development team in Switzerland, I would have to follow up and confirm their process, but I am guessing that it would be somewhat similar to what we do in the field. I will inquire!

As for adjusting the values, for the Liquid Verification Kit (LVK) process, it simply pauses after each cycle of transfers and allows you to open the LC database and make the changes directly in that and then resume (best to make changes to LC settings when paused!).

We do have an hsl library for modifying liquid class values during a method run though which you can download here. With this library, I have not had the need to use SQL to update/modify values in the LC database file. This is a custom library, so any feedback on performance is welcome!



I think DoE is a bit of overkill for most liquid class development situations.
I will generally adjust for precision and clean dispenses then for accuracy. I like tartrazine powder into the liquid to be optimized. For accuracy compare to a manually prepared standard curve.