How lab automation can shape the future of research
We need a more reliable way to reproduce and refine experiments, lab automation may hold the answers
A large part of biological research consists of repeating sets of tedious, error-prone, labour intensive tasks. For example a classical way to discover new drugs is to test if any compound in a very large library of molecules binds to a desired protein.
Given this fact there has been substantial efforts over the years to automate the more common processes of biological research. High throughput screening methods such as ELISA assays are often performed by liquid handling robots. A whole industry has sprung up around this lab automation efforts, with companies like Ginkgo bioworks, Opentron or the ECL all trying to make it easier to design and run experiments with robots. Much of the market is dominated by companies selling the actual hardware for this automation effort, such as TECAN or hamilton. The economical reality means that all those companies are tightly integrated with big pharmaceutical companies, since they are their main customers.
Because they rely financially on those large, conservative companies there is little drive in the lab automation community to foster open source models and tooling. In the current economic environment the newer, more innovative companies in the space have either gone out of business or slowly moved toward hiding their code away. For example Synthace, which started out with an open source language to define biological experiments, have completely given up on their efforts and removed their codebase from github. A notable exception to this rule is opentron, which have so far kept their lab automation code public.
The end result of this financial pressure to jealously guard their solutions is a net loss for both progress and long term economic returns. Classical lab automation robots like TECAN are hugely expensive and notoriously hard to code yourself, thereby ensuring that the consumer needs to pay expensive maintenance contracts to operate the hardware. More effort goes into obfuscating APIs and drivers, meaning that no third party ecosystems of solutions can develop to solve problems that the automation company might not have foreseen.
Contrast this with the vibrant state of the open source 3D printing community and it’s easy to see what lab automation companies are losing by falling into the trap of thinking building walled gardens makes more financial sense. It hints at a dated mindset inherited from big pharma, which heavily relies on monetising intellectual property.
Thinking that IP is the only way to develop financially is clearly disproven when one looks at software companies. Those are some of the world’s most valuable companies despite the fact that they routinely open source large parts of their codebases. For example, Google made the software which managed its large computational resources public with kubernetes.
Open sourcing lab automation software has incredible implications for reproducible research and a platform like versioned.science. This is because reproducing an experiment from a published protocol is not as straightforward as one might think. There is no standard way to describe a process, and information is often missing or unclear. This is what caused a much publicised reproducibility crisis in a wide array of scientific disciplines. The reality is that researchers often rely on protocols devised by their research groups since experienced peers are a much more reliable source of information.
The issues around reproducing research with error prone human to human communication goes away with a properly designed open source lab automation platform. The problem turns from trying to convey all aspects of a complex experimental process in natural language to simply running a script. Code is famously unforgiving in its ruthless idiocy; it will execute the steps described in your script exactly the same way every time regardless of their correctness.
Running and iterating upon an experiment can then be made much easier with a platform like github since describing it becomes a matter of editing code. It opens up a whole array of processes commonly available in software, such as a CI/CD pipeline which validates your experiment before you share it with others by merging it into a main branch.
This is what we think will form the core of the new kind of research which we will share with you on versioned.science. By including the code required to execute the experiments described in a research project we hope that reproducing research will become as easy as clicking a “run” button.