ScaaS: Science as a service (a research revolution)

CloudLab

Greetings, my fellow colleagues of tedium. “Give me your tired, your poor, your huddled masses yearning to breathe free …” Welcome to the new world of research: ScaaS.

It’s time to bring us out of the dark ages. No more old PCs running Windows 2000. No more equipment still using floppy drives. No more carpal tunnel syndrome from repetitive tasks a robot could (and SHOULD) do. It’s not just about our quality of life – it’s about changing the devastating trend of costly R&D.

I’m happy to say that the future is here – or, at least, we’re on the brink of it.  I’m calling it science as a service. It’s like software as a service, but for the research industry. Even if you aren’t familiar with the term SaaS, you’ve probably used an implementation of it. SaaS-based products host their software and your associated data on the cloud, and you interact with it via a simple web browser interface.

Now consider using that model for your favorite scientific experiment:

Instead of purchasing the hardware ($20k – $100k, or more), dealing with the software, maintaining the instrument, and doing the experiment by hand, you pop open your favorite web browser. You select the experiment and direct every detail by specifying a series of options (cell type, temperature, internal standard, instrument settings, etc. – if it can be altered, there should be an option for it). Maybe you also select parameters for how your data should be analyzed, how many times it should be repeated, or you select a desired completion date. Click, click, click, and your little experiment is on its way.* And you are on to bigger and better things.

Sounds great, right? Other people think so too. There are already a few names in this field. You should check out this great talk about how Emerald Therapeutics’ Symbolic Laboratory creates a construct for lean research (and how “lean research” could no longer be an oxymoron). TechCrunch blogged recently about Transcriptic and Benchling, and companies like Synthego, Gen9, and Gingko Bioworks are making headway too. 

So what’s keeping this from being immediately adopted in every lab in the country?

First, it’s probably because most experiments are not done in an automated fashion. If you go through a web interface to order an experiment, but then a human in a CRO does it for you, this doesn’t help much. It may save you some time, but it’s not a scalable or cost-efficient model. But just because these experiments aren’t normally done in an automated way doesn’t mean they can’t be done in an automated way. Most people still prefer grad students as a cheap form of labor (students making just over minimum wage, in fact), even though an automated instrument is more cost-effective in the long run.

But there’s another problem. It’s a mindset. Let’s face it: We scientists can be greedy. We just don’t want an experiment to be out of our hands. There is a biased attitude of trust that if you do it yourself, it’s “done right”. But the do-it-yourself model hasn’t worked out so well for us in terms of reproducibility. I’m not saying that you shouldn’t be concerned about handing over your experiments, but if a robot is doing the work you can take solace in that it will do what it’s programmed to do.** Robots don’t make complex errors like forgetting one element of a buffer recipe because it hasn’t had its morning coffee. The errors are standardized and if a major problem occurs, the robot stops entirely.

Lastly, scientists need to be more demanding about data. We need to make a priority of gathering it, storing it, and sharing it. In order to trust an experiment to a ScaaS system, the user needs to get back all the data (raw data, meta data, instrument files) they can get their hands on. Not just for the experiment at hand, but for the controls too. And not just the control run before their experiment – every control run. Ever. That should all be open-access. That way a user could investigate global changes in the behavior of the instrument, not just see an isolated period in time when their experiment was run. I would even suggest providing video records of the experiment in progress. (If you can afford a webcam to watch your kitty sleep all day, then a ScaaS center can afford them to watch their robots.) 

In conclusion – the solution is out there. But to adopt this system, we have to change the way we do science. We need to start integrating automation on all levels, incorporating computer science into the scientific way of life, and most of all, becoming gluttons for data.

* There is, of course, a potential problem with this. I’m not suggesting that only one powerhouse should dominate the market for a particular experiment. It would disastrous to find out later they’d done something wrong – remember the fiasco when we found out the major breast cancer cell line MDA-MB-435 was actually a melanoma cell line? I don’t think we need to get ourselves in that potential situation. I think competition within the private sector can do what it does best – have companies compete until a few “gold standard” options exist that are well-validated and trusted.

** The old adage “garbage in, garbage out” applies here. To properly program a robot, you have to have a unique blend of science and computer science savvy. I’m privileged to know some of these gurus, and in the right hands, this works.

How to buy new scientific equipment

Image

An acquaintance of mine started working in a new lab, and they have a beautiful, expensive, nearly brand-new liquid handler*. It’s currently unused and gathering dust. The technique they bought it to perform (which has to be laborious to do by hand, or they wouldn’t have purchased it in the first place) is currently being done. By hand.

*A liquid handler is a robot that is programmable to, literally, handle liquid. You can load it up with plates, tubes, reservoirs, petri dishes… whatever you’d like, and it will transfer liquid for you from one place to another. If you’d like your mind blown, check out my current favorite at HamiltonRobotics.com. Check out their youtube videos. (I’m not being compensated or anything to say this – it really is my favorite.)

The saddest part is that I’ve heard this song and dance before. A lab finally convinces their PI/Director/Money-Giver to purchase an automated system, and it sits there unused because… why? Most of the time only one person in the lab knows how to use it. Perhaps it worked for a while and then broke and you can’t get the supplier to fix it. It’s usually never used to its full potential, and in the meantime, people are still getting carpal tunnel from repetitive pipetting.

This post is very nuts and bolts, but I can’t help myself. I’ve decided to write a tutorial on how to buy a new piece of scientific equipment. Do I have credentials to speak on this topic? Yes. Have I been employed to sell scientific equipment for the last 10 years? No my friends. My education came from the streets. When you are put in charge of buying HPLCs, liquid handlers, peptide synthesizers, plate readers, mass spectrometers, and other things I can’t even remember, you learn a thing or two.

So here we go – these are my super-duper, number one, must-follow commandments to buying a piece of equipment. Share widely!!

1. Demo. Demo, demo, demo, demo, demo. Seriously. Do not purchase without a demo. Most manufacturers will offer to bring it on-site, but they’ll want to babysit you through one experiment and then leave. Ask them to leave it with you for a few days so you can find the bugs on your own (they are very good at avoiding them with a hand-held demo).

If they won’t bring it on-site, go to them if you can (but first, be wary – why do they want you to buy something without testing it?)

If they won’t bring it to you, and you can’t go visit them, but you absolutely have to have it, then write a conditional statement into your purchase agreement that allows for a trial period. Most manufacturers work on at least a net-30 basis if your lab has any credit (meaning you don’t have to pay until 30 days after it arrives). See if you can negotiate a 50% down net-30, and then 50% later (or even better terms). That way if it breaks or isn’t working properly, you still have leverage so they’ll pay attention to you and fix it.

2. Pretend you are dumb. I can’t tell you how many pieces of equipment could not actually perform the function they were built to perform. When they show you how it works, ask what every button means, and ask why they are using the software in a particular way. This is how you find the bugs. Let’s not live in a fairy tale here – there will be bugs. The sooner you find them, the sooner you can tell if you can live with them. Try to think about any variable you’d want to change during an experiment and see if the equipment can handle it. Be annoying, but in a charming way. Buy them a cup of coffee and express excitement for the product (while asking them every question you can think of).

3. Purchase equipment that exports raw data. For those of you who already analyze your own data – bravo, this is clearly a must for you. For those of you who don’t – why not? Are you sure you won’t ever need the raw data… ever? Even if you only use their analysis software, can you guarantee that the next software upgrade won’t change something that you can’t control? Most equipment will export raw data into a simple file, like a .txt, .csv, or .xml. You can probably find a way to work with whatever form it’s exported in (as long as it doesn’t come out in binary).

4. Modify your purchase agreement. If you were buying a new car, would you take the initial offer? Oh heck no. You’d negotiate. Why? Because you are a smart person, and you know the only time you have leverage is before the purchase is made. In this industry, most of the profits are made from the service contracts, not the equipment itself. That means they get more money from servicing their equipment than selling it. Don’t just sign what’s offered. Make a list of exactly what the equipment needs to do for you:

  • How often does it need to function?
  • What accuracy is required? (Be specific! Also, make sure you can test and confirm any of these numbers, and offer to share the data with them)
  • What does the software need to do?
  • What happens if the software upgrades? (Doesn’t it still need to do the things above? Why yes, yes it does.)

Include conditionals so that it has to do those things or you get to return it. You can ask your lawyer friends for some good legal terminology to use here.

Be prepared because they are going to push back a little on this. They will say they can’t guarantee their instrument will function perfectly all of the time. “Of course” – you say in a conciliatory tone, because you are buddies – “we’re in this together”. Ask them what they can guarantee. Make it clear that time lost is money lost. If it’s broken, how quickly can they fix it so it meets the previous expectations?

If you are going to hold them to a high standard, put yourself there as well. Get as much data as possible from your controls and experiments so you can show them when it works and when it doesn’t. Be meticulous and offer to share that data with them if you can.

5. Make them invested in you. Try to make them care about their instrument’s success in your lab. Perhaps you have friends who will all want to buy their product if it works well. Maybe you can use their equipment for a new application – this means more $$$ for them, and you can offer to work with a product manager to write a new application note. Or maybe you can just get them to care about you and your work, and the negative impact that’s made when their equipment doesn’t function properly. This is why you have to be friendly and charming through this whole process – this is a dual investment, and they should want to work with you.

Hungry for more? If you’ve read all this and want to keep going, then we should be friends. Really. See my super pro tips below:

Super pro tip #1: Buy your own computer. If the equipment needs to use a computer, purchase it yourself (or better yet – build it). Ask what types of connections are required to hook up to the instrument (Serial port? Ethernet? PCI card?). This way you are in control if the computer breaks down, you can get a decent model to your specifications, and you’ll save some moolah. Otherwise they’ll charge you $3000 for a crappy Dell laptop. (No offense to Dell, but seriously).

Super pro tip #2: Ask about software compatibilities. If it only runs on Windows XP, be afraid. Be very afraid. If getting stuck with a $3k Dell laptop is bad, getting stuck with one only running XP for the next 5 years is worse. This will help you gauge how much they care about their software. In my experience, most instrumentation companies are made of hardware people, not software people. The hardware can function beautifully but bad software will screw it all up.

That’s it! If you have any additional tips or comments, let’s hear ’em. We should be fighting to make research better, not putting up with the same problems over and over again.