Week 3 Review: Lab Automation

Week 3: Lab Automation

HTGAA 2026 — Fiona Connolly

What Lab Automation Can Do for Us?

Lab automation is simply automating the processes in the lab. Scripted protocols, and integrated instruments to carry out experimental procedures with ideally minimal manual intervention. Particularly in molecular biology, this typically translates to very precise , temporally and temperature controlled liquid handling across the scale from picoL to Litres. The precise transfer of reagents, cultures, or genetic constructs between wells, plates, and vessels.

Automation is useful for lab processes in speed and consistency, a protocol run on a automated system produces the same volumes, timings, and positions every time, removing operator-dependent variability that limits reproducibility in manual workflows.

These systems range from large integrated platforms (Hamilton STAR, Beckman Biomek, CloudLab, DAMP Lab) down to benchtop robots accessible to academic labs,like the OT2 and CyBio Felix. One that we explore further this week is the OpenTrons OT-2, an open-source liquid-handling robot that costs roughly $5k (compared to $100k+ for legacy systems) and is programmed entirely in Python. It offers ±0.1 mm positional accuracy and <1% volume CV at the microlitre scale, making it practical for most molecular biology, microbiology, and synthetic biology workflows.

Here is a typical OT-2 deck setup for a combinatorial screening experiment:

PLACEHOLDER ![OT-2 deck layout showing tip rack, source plate, destination plate, tube rack and pipette head]

Example Application: Combinatorial Cross-Culture Screening

To illustrate what precise automated pipetting enables, here us common synthetic biology task: EXAMPLE HERE

PLACEHOLDER ![Heatmap interaction matrix showing 12 sender strains crossed against 32 receiver strains]

The practical difference between manual and automated execution at this scale:

Q1: Published Paper Using Automation for Novel Biology

Towards Automation of the DBTL Bioengineering Cycle: Application to Testing and Characterization of Standard Bioparts

Pushkareva, A., Beltran, J., Díaz-Iza, H., Arboleda-García, A., Boada, Y., Vignoni, A., Picó, J. (2023). XLIV Jornadas de Automática, 453–458. DOI: 10.17979/spudc.9788497498609.453

Pushkareva et al. address a gap in the Design-Build-Test-Learn (DBTL) cycle: while much prior work has automated the Design and Build steps individually, few efforts have tackled the Test and Learn steps together. This paper presents an integrated automated workflow combining the Opentrons OT-2 liquid-handling robot with an Agilent Biotek Cytation 3 plate reader to systematically characterise standard genetic bioparts.

The automated Test step worked as follows. The OT-2 (fitted with a Multichannel P300 and Single Channel P1000) performs a two-part protocol controlled via Jupyter notebook. -First, it dilutes 7 bacterial culture samples 1:4 in M9 minimal media with glucose, then transfers them to the plate reader for an initial OD600 measurement. -The OD values are fed into a template spreadsheet that calculates the volumes needed to normalise all cultures to OD 0.1. —-The OT-2 then executes a second protocol using those calculated volumes, producing a standardised 96-well plate that goes into a 16-hour incubation/measurement experiment at 37°C and 230 rpm, recording both absorbance (600 nm) and fluorescence (530/488 nm).

PLACEHOLDER ![Automated Test and Learn workflow: OT-2 dilution and normalisation, plate reader measurement, parameter identification and cross-device prediction]

For the Learn steps, -they used the resulting calibrated dataset (particles, MEFL, MEFL/particle, growth rate) from two GFP expression constructs — a low-copy (pSC101, C_N=5) and high-copy (ColE1, C_N=35) plasmid, both with identical promoter (BBa_J23106) and RBS (BBa_B0030). -Using a growth-independent protein production model and genetic algorithm-based parameter identification on the low-copy data alone, they showed that the same parameter values could accurately predict the protein production of the high-copy device simply by changing the copy number, with comparable prediction error (MSE ~3.47×10⁸) to the optimisation error (MSE ~3.50×10⁸).

This is great example of using automation to scale up a standardised experiment and generate invaluable data because it demonstrates a practical closed loop: automated liquid handling produces consistent enough data that the Learn step (model fitting and prediction) actually works across devices. The reproducibility enabled by the OT-2 is what really makes this type of experiment doable and reprodicible.

Q2: Automation Plan for Final Project

Project Overview

The goal of on of final project ideas is to develop cell-free colorimetric biosensors that produce a visible colour change when a target biomarker exceeds a clinically relevant threshold. Automation would create a path for screening a combinatorial library of genetic circuit components (promoters, RBS variants, reporter genes, sensor elements) to identify which constructs give the best dose-responsive colorimetric output for each target analyte.

Beyond single-analyte detection, I want to explore the feasibility of multiplex circuits that detect more than one biomarker in a single reaction, using orthogonal detection modalities (e.g. toehold switches for RNA targets, transcription factor-based circuits for protein targets, and CRISPR-Cas12a for DNA targets).

I am planning for two application areas, each with three target biomarkers.

Two Use Cases and Their Molecular Targets

PLACEHOLDER ![Biosensor target panel showing pathogen exposure markers and cancer recurrence biomarkers with detection modalities]

Use Case 1 — Pathogen Exposure Markers:

The first application targets three infectious disease biomarkers relevant to resource-limited settings where point-of-care colorimetric diagnostics would have the most impact. For Ebola, the target is the secreted glycoprotein (sGP), a decoy antigen present in infected blood that can be detected earlier than PCR-based methods using sandwich immunoassay or CRISPR-Cas13a approaches targeting viral RNA. For HIV, the target is the p24 capsid antigen, detectable at approximately 10 pg/mL by colorimetric ELISA with ultrasensitive methods reaching 0.5 pg/mL. For tuberculosis, the target is lipoarabinomannan (LAM), a mycobacterial cell wall glycolipid excreted in urine at a median concentration of approximately 137 pg/mL in TB-positive individuals, making it suitable for non-invasive sample collection.

The multiplexing opportunity here is that these three targets are molecularly orthogonal (two proteins and one glycolipid), so independent detection channels could in principle operate in the same cell-free reaction without crosstalk.

Use Case 2 — Cancer Recurrence Biomarkers:

The second application targets three biomarkers associated with cancer recurrence monitoring. Circulating tumour DNA (ctDNA) carrying EGFR mutations (L858R, exon 19 deletions) is detectable at variant allele frequencies as low as 0.02% using CRISPR-Cas12a or multilevel toehold switch circuits, and predicts progression in approximately 64% of NSCLC cases before clinical detection. Circulating microRNAs miR-21 and miR-155 are upregulated 1.5–1.7-fold in the plasma of patients with breast, colorectal, pancreatic, and liver cancers, and are detectable without amplification using gold nanoparticle aggregation-based colorimetric assays. Exosomal PSA tracks prostate cancer recurrence at a threshold of 0.2–0.5 ng/mL, with exosomal urine tests achieving 92% negative predictive value.

These three targets span DNA, RNA, and protein modalities, which again creates the possibility for orthogonal multiplex detection in a single reaction format.

What Needs to Be Automated

For each use case, the screening task is the same: test a library of genetic circuit constructs against an 8-point concentration gradient of the target analyte, measure the colorimetric and fluorescent output over time, and identify which constructs produce a clear dose-response curve with a visible colour change at or below the clinical threshold concentration. The construct library for each target would include 3–5 promoter strengths × 3 RBS variants × 2 reporter genes (LacZ for colorimetric, mCherry for fluorescence), giving 30–60 constructs per target. Across 6 targets, that is 180–360 constructs to screen, each against an 8-point gradient — roughly 1,500–3,000 individual reactions. This is not feasible manually.

I have planned the automation for two scenarios depending on available infrastructure.

Scenario A: OT-2 + Standard Benchtop Automation

PLACEHOLDER ![OT-2 benchtop automation workflow for construct screening with dose-response readout]

In this scenario, the entire DBTL cycle runs on an OT-2 with an attached plate reader (Biotek Cytation or similar), following a protocol structure similar to Pushkareva et al.

Build: The OT-2 assembles constructs via Golden Gate or Gibson assembly into 96-well format, transforms into competent cells, and plates for colony selection. After overnight growth and colony picking, constructs are arrayed in a source plate.

Test: The OT-2 distributes cell-free protein synthesis (CFPS) master mix into a 96-well plate, adds DNA constructs from the source plate, then sets up an 8-point serial dilution of the target analyte across columns. The plate is sealed and incubated at 37°C for 1–6 hours, then read for absorbance (570 nm for colorimetric reporters) and fluorescence (530 nm for mCherry). The plate layout allocates rows A–F to 6 constructs, columns 1–8 to the analyte gradient, columns 9–10 to no-analyte controls, and columns 11–12 to positive controls, with row G as a blank and row H for calibration standards.

Learn: Dose-response curves are fitted to a Hill function for each construct. Constructs are ranked by dynamic range above the clinical threshold, signal-to-noise ratio at the threshold concentration, and time to visible colour change. Top performers from each target are carried forward into the next DBTL iteration.

At 96-well scale, each OT-2 run screens 6 constructs against one target. Screening the full library of 30–60 constructs per target requires 5–10 plates per target, or roughly 30–60 plates total across both use cases. At one plate per run (including setup and incubation), this takes approximately 2–3 weeks of daily runs.

Example OT-2 protocol for the analyte gradient step:

from opentrons import protocol_api
 
metadata = {'apiLevel': '2.13',
            'description': 'Biosensor dose-response screen — CFPS colorimetric'}
 
def run(protocol: protocol_api.ProtocolContext):
 
    # ── Labware ──────────────────────────────────────
    tiprack_300 = protocol.load_labware('opentrons_96_tiprack_300ul', '1')
    tiprack_20  = protocol.load_labware('opentrons_96_tiprack_20ul', '4')
    cfps_reservoir = protocol.load_labware('nest_12_reservoir_15ml', '2')
    construct_plate = protocol.load_labware('corning_96_wellplate_360ul_flat', '3')
    dest_plate  = protocol.load_labware('corning_96_wellplate_360ul_flat', '6')
    analyte_rack = protocol.load_labware(
                     'opentrons_24_tuberack_eppendorf_1.5ml_safelock_snapcap', '5')
 
    p300 = protocol.load_instrument('p300_multi_gen2', 'left',
                                     tip_racks=[tiprack_300])
    p20  = protocol.load_instrument('p20_single_gen2', 'right',
                                     tip_racks=[tiprack_20])
 
    # ── Step 1: Distribute CFPS master mix (10 uL/well) ──
    p300.distribute(10, cfps_reservoir['A1'],
                    dest_plate.wells()[:72], new_tip='once')
 
    # ── Step 2: Add DNA constructs (2 uL each, rows A-F) ──
    for row in range(6):
        for col in range(12):
            p20.transfer(2,
                construct_plate.wells()[row * 12 + col],
                dest_plate.wells()[row * 12 + col],
                new_tip='always')
 
    # ── Step 3: Analyte serial dilution (cols 1-8) ────────
    #    Concentrations: 0, 0.1, 1, 10, 100, 500, 1000, 5000 pg/mL
    analyte_vols = [0, 0.2, 0.5, 1.0, 2.0, 4.0, 6.0, 8.0]  # uL from stock
    buffer_vols  = [8.0, 7.8, 7.5, 7.0, 6.0, 4.0, 2.0, 0]   # uL buffer
 
    for col_idx in range(8):
        col_wells = dest_plate.columns()[col_idx][:6]  # rows A-F only
        if buffer_vols[col_idx] > 0:
            p20.distribute(buffer_vols[col_idx],
                           analyte_rack['A1'], col_wells, new_tip='once')
        if analyte_vols[col_idx] > 0:
            p20.distribute(analyte_vols[col_idx],
                           analyte_rack['B1'], col_wells, new_tip='always')

Scenario B: Ginkgo Bioworks Cloud Lab / DAMP Lab

PLACEHOLDER (![Cloud lab automation workflow using Ginkgo Nebula or DAMP Lab integrated instruments]))

In this scenario, the entire workflow is submitted remotely and executed on integrated high-throughput instruments at Ginkgo Bioworks (via Nebula) or the DAMP Lab at Boston University.

Build: Construct designs are submitted digitally. Gene synthesis is handled by Twist or IDT. The Echo 525 acoustic liquid handler transfers construct DNA at nanolitre precision into 384-well plates, and a Bravo liquid handler stamps in additional reagents.

Test: A Multiflo dispenser adds CFPS lysate to all wells. The plate is sealed (PlateLoc), incubated at 37°C (Inheco), unsealed (XPeel), and read on a PHERAstar plate reader for absorbance, fluorescence, and luminescence. The entire sequence runs without manual plate handling.

Learn: Data is exported via API for automated dose-response fitting, Hill coefficient extraction, and LOD calculation. Constructs are ranked by the same criteria as Scenario A.

The key advantages of the cloud lab over benchtop are scale and precision. At 384-well format, a single plate screens 24 constructs against one target (4× the throughput of OT-2). The Echo transfers at nanolitre precision, reducing reagent consumption. The integrated seal/unseal/incubate workflow eliminates manual plate moves entirely. And all 6 targets can be run in parallel, completing the entire screen in approximately 2–3 days rather than 2–3 weeks.

Cloud lab protocol pseudocode (per target):

1. Echo 525:    Transfer construct DNA (2.5 nL each) into 384-well plate
                → 24 constructs × 16 replicates per plate
2. Echo 525:    Transfer analyte at 8 concentrations across columns
                → nanolitre serial dilution, no tip waste
3. Bravo:       Stamp CFPS reagent master mix into all wells
4. Multiflo:    Dispense CFPS lysate to start protein expression
5. PlateLoc:    Seal plate
6. Inheco:      Incubate 37°C, 1–6 hrs
7. XPeel:       Remove seal
8. PHERAstar:   Read absorbance (570 nm) + fluorescence (530/488 nm)
                → kinetic or endpoint, per experimental design
9. Data export: Dose-response curves → Hill fit → rank constructs

Multiplex Circuit Design Considerations

For both use cases, the longer-term goal is to combine the top-performing single-analyte sensors into a multiplex format where two or three biomarkers are detected in the same cell-free reaction. This is feasible because the targets within each use case are molecularly orthogonal: in the pathogen panel, sGP (protein), p24 (protein), and LAM (glycolipid) can each bind to distinct sensor elements; in the cancer panel, ctDNA (nucleic acid), miRNA (nucleic acid), and exosomal PSA (protein) can be read by CRISPR-Cas12a, toehold switches, and aptamer-based circuits respectively. Each sensor module would drive a spectrally distinct reporter (e.g. LacZ/yellow at 405 nm, mCherry/red at 587 nm, and a luciferase/blue for luminescence), allowing deconvolution on a standard plate reader.

The automation requirements for multiplexing are the same as for single-analyte screening, but the number of conditions increases: each multiplex combination needs to be tested against a matrix of analyte concentrations (rather than a single gradient), making the cloud lab scenario (Scenario B) strongly preferable for this stage.

Comparison of Scenarios