AI Brings the Next Generation of the Electronic Lab Notebook

The initial benefit [of ELNs] was documentation, providing a system of record when you did an experiment. But the real benefit comes when you get that data back out.

Say that I, as a scientist, need to do something. With an ELN, I can easily see if there is experience in the company of doing something like this before. This is a transformational difference from keeping paper records, where I couldn’t know that. So now I’m doing better experiments, and the company is saving money because we won’t end up duplicating things.

The other thing that has really changed is that the latest ELNs are much more about collaboration. That could mean collaboration as in, “there are two of us in a lab and we both need to do something,” all the way up to cases where maybe you have commissioned a contract research organization to do something and you want to stay actively involved and keep that as an interactive collaboration between scientists, as opposed to just providing a bunch of work and a chunk of money. 

Whether it is Sapio’s ELN or an ELN made by anyone else in this room, one of the real barriers for scientists is learning how to use the software. Now, it’s always true for software that you will need to learn new things, but for ELNs, these are complex scientific tools; there is a meaningful barrier in learning it and getting up to speed. You can do something in an ELN, spend a week in the lab, and then when you come back to the ELN, you can’t remember how to work that process again.

What ELaiN does is flip this on its head. You don’t have to learn how to tell the ELN what to do – you just intuitively express what you’d like to do, and something else can figure out how that would work in the software. The learning barrier has been reduced to as close to zero as you can get, short of having the ELN read your mind.

We’ve said already that ELNs are good because there is a ton of really good information about what has happened in the past that can help inform the future. But right now, it is completely passive. As a scientist, I have to go to the ELN and research what we’ve done before, then maybe I go and talk to the project team, then the computational experts to help me decide what to do, and then I go back into the ELN. It’s completely out of the loop in helping me.

With ELaiN, if the AI can understand the experiment – not just looking at words on the page, but understanding the scientific meaning of that experiment – then it can know, if you want to make a certain molecule, here’s how to do it. Or if you want to optimize that sequence, it would know what a scientist would do. If it can do those things, then it becomes an active co-scientist. It’s like being a chemist with another chemist always looking over your shoulder. Or being a molecular biologist with a bioinformatician available to you at all times.

This is why we are saying it’s a third-generation ELN, because it is solving these two fundamental problems. This will mean that it is a lot easier to adopt and scientists can be much more innovative as they use it – better experiments, maybe also fewer experiments, but better and faster scientific progress.

Whatever a scientist would do with the help of past data and computational experts, or any question you would ask of the internet or the scientific community, you should be able to ask the ELN and get a high-quality, well-informed, scientific answer. And the key thing is, it doesn’t interrupt your workflow.

Say I’m in the ELN, I’m going to make these molecules, and I think, “I'd better check there isn’t a toxicity risk first.” Then I have to stop and go and ask somebody. That interrupts your flow. If I can be in the ELN and have this realization and then three seconds later I have a yes/no answer, I haven’t interrupted my chain of thought or my workflow and I still get those answers. The research experiment is going to be much more fluid; research at the moment is very piecemeal. You are always switching between doing something in the software, doing something in the lab, doing something with colleagues, doing something with instruments…

The other thing is, if you think about the junior scientists that are coming into the industry, they’ve grown up with smart assistants. Then they get into the lab and they are presented with software from 10 years ago where you have to fill in forms and do X, Y and Z. They must be horrified, or at least, confused. You are teaching people who have just come out of university how to behave in a lab as if it were still 20 years ago. It’s much more efficient to be able to present them with something that they are already familiar with from their daily lives and just continue it into science. 

There is still a lot of concern about security. I think we have good answers to addressing that security, but the first reaction of a lot of people is, “You are taking my questions and sending them to Meta or Google or Grok!” But no, it's all contained within a secure place and it’s all validated in the same way. But security is certainly a key talking point.

Consistency is the other one. There is a lot of concern about perhaps not getting a consistent answer [from scientific AI assistants]. Every time we do something in ELaiN, like a search, it doesn’t just show you the answer with no context – it will show you how it did that search. You can go in and read that and know what exactly you are looking at because you have been given the context. Say, for example, you asked how to make a certain molecule and ELaiN brought up some reactions. There is a whole audit trail in there that reports which algorithms were called and where these suggestions came from. You have to show the scientists the working, then they can understand the process and accept the answer.

Another point is, even though it is using a large language model and therefore is variable, ELaiN is calling our application programming interface (API), which is deterministic. So as long as it gets the right command, you will get the right answer. And if you send the same input, you will get the same answer every time. 

Crucially, it is a scientific assistant. The scientist is responsible for the experiment and is responsible for ensuring that the experiment is scientifically valid, or the analysis is valid, or the template that an ELN has generated from a document is valid. [AI] doesn’t absolve the scientist of any responsibility, but it does alleviate some of the grunt work of getting there. Scientists should absolutely be reviewing these outputs and saying, “this heatmap doesn’t make sense,” for example, but this is easy to fix.

As a scientist, I might spend an hour reviewing a [generated] template and fixing a few things. But if I wanted to build such a template from scratch using a 20-plus-page SOP, at two hours in, I would still be reading the SOP and understanding what it meant before I could even start.

I think, firstly, this is a relatively straightforward and practical application of AI. That should hopefully help many people get over the hump of thinking “is AI a help or a threat?”, because clearly in this case, it is providing you help.

People more and more are using ChatGPT or Claude or other AI models more often in their daily life, whether that is to research something on the internet or to plan a holiday. People have built up that experience. When they can come to their lab notebook and have that same experience, they will be increasingly more comfortable with that. This is a very practical and simple introduction to AI, as well as [an introduction] to the other things that everybody is trying to do with AI at the moment.

The best training or advice [for using ELaiN] would be, just ask your questions as if you are talking to a colleague.