Troubleshooting an Automated ELISA Assay
Using AI-Assisted Analysis

The Challenge

An automated ELISA workflow involving serial dilutions, multiple sample and control types, robotic liquid handling, automated washing, plate reading, and downstream potency analysis was expected to produce a clear dose-response curve with acceptable replicate agreement and signal-to-noise performance.

Instead, several problems kept appearing during development:

• Top optical density lower than expected
• Background signal varying between runs
• Recurring abnormal behaviour in certain plate regions
• Inconsistent replicate agreement
• Occasional positive control failures
• Small liquid handling changes producing unexpectedly large assay effects

The difficult part was that the assay was not simply broken. It was often close to acceptable — which made troubleshooting significantly harder. When an assay is only just failing QC, every variable can look plausible, and it is easy to overadjust or miss the real cause.

The Approach

The troubleshooting process combined wet lab reasoning with structured AI-assisted investigation. The assay was broken down into its operational components rather than treated as a single instrument fault:

• Plate layout and replicate structure
• Dilution series design
• Robotic transfer steps and liquid class settings
• Reagent addition order and direction
• Pre-dispense and post-dispense behaviour
• Washer performance
• Incubation timing and conditions
• Plate reader outputs
• Run-to-run QC trends

AI was used as a reasoning and analysis layer to help compare runs, organise hypotheses, identify recurring spatial patterns, and prioritise the next most useful experiments. It did not replace scientific judgement — it acted as a structured troubleshooting assistant, holding multiple variables in view simultaneously.

Liquid Handling Method Review

The assay required accurate handling of small volumes, serial dilutions, viscous or detergent-containing reagents, and repeat dispensing across plate columns. Liquid class settings were reviewed and adjusted, including:

• Aspirate and dispense speeds
• Pre-dispense and post-dispense volumes
• Air gaps and conditioning volumes
• Breakoff behaviour and multi-dispense strategy
• Reagent-specific handling differences

Particular attention was given to the difference between dispensing the correct volume and achieving biologically reliable assay behaviour — two things that can look identical at the method level but produce very different results at the plate level.

Plate and Reagent Effects

The investigation assessed whether reagent addition direction, timing, multi-dispense behaviour, or first-column effects were contributing to abnormal wells or edge-related issues. Patterns in the plate data were reviewed across multiple runs to determine whether failures were random or spatially consistent.

This distinction matters: repeatable spatial patterns point to workflow-driven causes, while random failures suggest reagent stability, mixing, or environmental variables.

Signal and Background Investigation

Rather than assuming a single cause for the low top OD and variable background, the investigation considered the full signal chain:

• Conjugate handling and concentration
• Substrate and stop solution addition
• Wash consistency and carry-over
• Incubation conditions and timing
• Plate lot effects
• Reader settings and calibration
• Reagent stability and dilution accuracy

This broader view helped avoid the common trap of over-focusing on one instrument or one parameter too early in the troubleshooting process.

Consumable Impact

One key finding was that not every failure mode comes from automation parameters. A new plate lot restored expected top OD performance in one case, demonstrating that consumables could significantly affect assay behaviour independently of liquid handling settings.

This reinforced the importance of troubleshooting the full system — not assuming the robot is always the source of the problem.

Role of AI in the Troubleshooting Process

AI was used to support the troubleshooting process in several practical ways:

• Comparing assay runs and development notes across a long timeline
• Tracking which method variables had changed between runs
• Identifying recurring plate-position effects
• Helping interpret dose-response and signal-to-noise behaviour
• Generating structured troubleshooting hypotheses
• Prioritising experiments that could separate competing causes
• Assisting with documentation of findings and decisions
• Explaining liquid handling behaviour in plain language for method review