Week 3 Lab: Lab Automation

Opentrons Designs

I tried to push Opentrons to the limit and chose a fairly hard design: the Mitsudomoe, a traditional Japanese family crest (Kamon) associated with my family. The result didn’t come out particularly clean, but with higher resolution and non-sequential pipetting (for speed) it would be more tractable.

Reference design and Opentrons version:

Lab Automation

Find a published paper that uses Opentrons or another automation tool for a novel biological application.

The paper I chose is AssemblyTron: flexible automation of DNA assembly with Opentrons OT-2 lab robots, by Bryant et al., published in Synthetic Biology (2023). The authors developed an open-source Python package called AssemblyTron that connects j5 DNA assembly design software to an Opentrons OT-2 liquid handling robot, allowing users to go from a digital DNA design to a physically assembled construct with minimal hands-on work.

What makes this paper compelling is that it automates the entire Build step of the Design-Build-Test-Learn cycle, traditionally the most manual and error-prone part. AssemblyTron handles PCR setup (including calculating optimal annealing temperature gradients), DpnI digestion, and final multi-fragment assembly on the OT-2. The authors validated the system by performing Golden Gate assemblies and in vivo assemblies of four-fragment chromoprotein reporter plasmids, achieving fidelity comparable to manual assembly. They also demonstrated automated site-directed mutagenesis. The key takeaway is that affordable, open-source automation can make DNA assembly more reproducible, less wasteful, and accessible to labs without expensive biofoundry infrastructure.

What I intend to do with automation tools for my final project.

In general, I want to use my adaptive AI system for scientific discovery at a small scale, something realistic as a final project given the resources we have from Twist and Ginkgo Bioworks.

Idea 1: Promoter design for maximum expression. I would order oligos from Twist, clone them into reporters, and observe expression in E. coli. Fluorescence intensity would be the reward signal. Two rounds may be feasible.

Idea 2: In silico validation only. The most feasible version is to ditch the lab-in-the-loop entirely by performing validation in silico. This also allows much more complex protein designs since we are no longer constrained by what is physically feasible to test on the project budget.

Idea 3 (the dream version, not feasible in this timeframe). Use the system to discover higher-order transcription factor combinations that forward-program iPSCs into a target cell type. The computational engine uses Bayesian optimization to predict TF combinations, balancing exploration and exploitation based on experimental results. To handle the cloning overhead, I would outsource synthesis of polycistronic lentiviral transfer vectors to Ginkgo Bioworks’ Nebula platform, which algorithmically assembles the DNA and returns plasmids in a 96-well format. Each vector can carry 3 to 4 TFs linked by 2A peptides, and co-transduction with multiple vectors allows testing of even larger combinations.

The OT-2 would automate lentivirus production by dispensing transfection reagent into arrayed HEK293T packaging cells, harvesting viral supernatant, and transducing iPSC cultures. The robot would also handle the post-transduction media schedule. Because lentivirus integrates into the genome, TF expression is sustained throughout the differentiation window without repeated dosing. At the endpoint, high-content phenotypic imaging quantifies differentiation efficiency in each well, and the data feeds directly back into the Bayesian model to predict a more refined batch of TF cocktails for the next automated run.