Aquarium: Lab in the Cloud




Standarization and reproducibility are major problems in the field of Synthetic Biology, and other Bio-related fields. In the Klavins lab, we designed and implemented a product called Aquarium to formalize the wetlab process. Aquarium is an environment for specifying workflows and managing laboratoy operations in an easy-to-share, computer encoded, human and machine interpretable manner.

The system consists of a programming language for protocol and workflow specification, a library of tested workflows including a self-booting protocol for easy-installation in the lab, a laboratory management system, and an online repository. It's a lab in the cloud - allowing designers to submit genetic constructs to be built, specific ways to build them if necessary, and receive data without ever having to step foot in a wetlab.

Aquarium LLC was spun off in 2015, and is today the UW Biofabrication Center (UWBIOFAB). Check out our fun commercial for the BIOFAB below (written and directed by me - and starring two coworkers!)

T. Amimeur, Y. Yang, D. Younger, M. Gander, C. Takahashi, A. LaMarca, J. Scofield, D. Fox, J. Bilmes, E. Klavins. Aquarium: Programmable Wetlab, Intel Science and Technology Center for Pervasive Computing Conference (ISTC-PC), 2014.






Wetlab Object Tracking and Protocol Checking



Aquarium was developed as a part of the Intel Center for Pervasive Computing capstone project. We worked with the center to set up numerous cameras at each lab bench, and over the course of a year, collected over 18 TB of video data. Our machine learning collaborators at the ISTC will use this data for a variety of research topics.

As a part of Aquarium's goal of reproducibility, I used a portion of this video data to write object-tracking software for lab equipment usage during protocol execution. I wrote an Aquarium plug-in in OpenCV that uses the SIFT (scale-invariant feature transform) algorithm to perform protocol-checking - to minimize contaminitaion and process variability during genetic circuit construction.

T. Amimeur, E. Klavins. Aquarium, Intel Science and Technology Center for Pervasive Computing Conference (ISTC-PC), 2015.





Beluga: Automated Genetic Circuit Design


* IWBDA 2015 Best Poster Award *


Our focus with Aquarium was to automate and improve the building and testing of genetic circuits. Following Aquarium's transformation into the BIOFAB, I became very interested in automating the design portion of genetic circuits.

I spent the remainder of my thesis working to develop Beluga - a decision-theoretic agent for recommending optimal genetic designs under experimental and biological uncertainty. Beluga is a framework for iteratively exploring and learning about a space of genetic designs in a goal-oriented manner. It performs Bayesian inference on the parameters of a design space directly from experimental data to inform the next set of designs to be characterized.


Beluga + A.I.

In another iteration of Beluga, I approached finding an optimal genetic circuit as a non-deterministic search problem, where the agent must simultaneously learn about its environment and reach a desired goal. I presented an implementation of Beluga as a Partially Observable Markov Decision Process (POMDP) at IJCAI's Workshop on AI in Synthetic Biology.



T. Amimeur, E. Klavins. Automated Experimentally-Driven Design of Genetic Circuits, International Workshop on BioDesign Automation (IWBDA), 2015.


T. Amimeur, E. Klavins. MDP-based Planning for Design of Gene-Repression Circuits. Proc. of the IJCAI-2016 Workshop on AI in Synthetic Biology, 2016.



Creative Commons License
Beluga by Tileli Amimeur, Klavins Lab, Dept. of Electrical Engineering at the University of Washington is licensed under a Creative Commons Attribution 4.0 International License.