Automating Sample Prep with Liquid-Handling Robotics

Team Dynamic Matrices | 30 September 2025

The most common automation mistake we see is treating a robot as a faster technician. Moving a manual protocol onto a liquid handler does not improve it. If the protocol was inconsistent by hand, the robot will execute it inconsistently, at scale, with nobody watching. Automation does not fix an inconsistent protocol. It reveals one. Here is how we approach the transfer, from what we have learned automating our own matrix and capture workflows.

The undocumented steps

A skilled technician makes dozens of adjustments no protocol records: a pause after aspirating a viscous reagent so the tip fills completely; a change of angle to avoid bubbles; a visual check that a pellet resuspended; a second flick of a tube. None of that transfers. The robot does what the method file says, and the method file was written from the protocol, not from what the technician actually did.

Before automating, watch the person who gets the best results run the protocol, and write down everything they do that is not in the SOP. Those are the steps that have to be engineered into the method.

Liquid classes: where hydrogels, cells and lysates go wrong

Default liquid classes on every platform assume water. Three of the materials in 3D workflows do not behave like water.

Viscous hydrogel precursors. Slow to aspirate, slow to dispense, prone to air gaps and retained volume in the tip. A default class under-dispenses and introduces bubbles that ruin imaging. Fix: reduce aspiration and dispense speeds, add a post-aspiration delay, use wide-bore or low-retention tips, and consider reverse pipetting for the most viscous reagents. A matrix that is low-viscosity at room temperature (which is how we formulated DyNAtrix) reduces the problem at the source.

Cell suspensions. Cells settle. Over a 384-well plate run of several minutes, the first wells get more cells than the last. Fix: measure the settling rate of your cell type in your medium (it varies from under a minute for large aggregates to several minutes for single cells) and build in timed resuspension mixes at intervals shorter than that. Active mixing accessories (orbital shakers, magnetic stirrers in the reservoir) help for long runs.

Complex lysates. Viscous from DNA, heterogeneous, foamy with detergent. Fix: a nuclease or shearing step before transfer, low-retention tips, slow speeds, and liquid-level sensing to handle volume variation.

Validate gravimetrically

For every liquid class and volume you will use, dispense into pre-weighed tubes or plates and weigh. Convert mass to volume with the reagent's density. Target accuracy within 2 percent and precision (CV) within 1 percent for critical volumes. Do this for every new reagent, every new tip lot, and after every service. It takes an hour and it is the only way to know the robot is dispensing what you think it is. ISO 8655 sets out the procedure for pipettes and applies equally to automated channels.

Designing for walk-away reliability

Accuracy is necessary and not sufficient. A method that is accurate for 95 wells and fails on the 96th because a tip clogged has to be caught. Build in:

• Liquid-level sensing on aspiration, so an empty reservoir or a clot is detected
• Dispense verification where the platform offers it (pressure-based or capacitive)
• Error handling that pauses and alerts rather than continuing through a failure
• Full unattended trial runs before production, with nobody in the room, because the failures you find with someone watching are not the ones that happen at 2 a.m.

Automate selectively

The temptation is to convert the whole protocol end to end. The better result usually comes from automating the steps that are high-precision, repetitive and error-prone by hand (serial dilutions, plate replication, reagent additions to a full plate, fixed-time incubations) and leaving steps that require judgement or are cheap by hand where they are, until the chemistry is redesigned to remove the judgement. A protocol with three automated steps that never fail beats one with fifteen that fail weekly.

Where the chemistry has been designed for it, that boundary moves. A matrix that dissolves in a fixed time on a defined trigger, and a capture step that is an addition, an incubation and a spin, contain no judgement steps to begin with. That is the design principle behind both of our products, and it is why they automate as additions rather than as workarounds.