Thread by @s_r_constantin

Yikes!

I was rooting for them to succeed, but this report has some damning customer quotes about their machines not actually working.

https://twitter.com/ScorpionFund/status/1438129523678138371

It’s a very rhetorically aggressive report and makes some unfair arguments.

Saying that “just” miniaturizing assays and running them faster doesn’t make a difference is ignorant. Moore’s Law is “just” making smaller features and faster computer chips.

But unless the quotes from unhappy customers are fabricated, this is pretty unanimous disappointment in the machine’s price, accuracy, reliability, and usability from 14 major biotech/pharma companies.

The core tech of Berkeley Lights is a novel method to move single cells around with light.

This is cool! But why is it better than moving single cells around with droplets and electrodes (the standard for single cell sequencing?)

Given that multiple customers are complaining that BLI’s machine does the same thing as a FACS machine for 4x the cost and more equipment failures, that says they *aren’t even using it* for single-cell applications but rather bulk cell counting/sorting.

In which case, I don’t know how they managed to sell $2M machines to Pfizer and Bristol Myers Squibb for that use case, but that’s nuts. You don’t need expensive, buggy experimental tech to count and sort cells.

The real wtf moment was this, from a former BLI scientist: “Instrument is based on chips that contain silicon, and cells don’t behave naturally against a silicon surface; company never thought through the problems it creates.”

If they're letting the cells touch silicon directly, this is a VERY OBVIOUS bad. Silicon is not biocompatible. You have to coat it with some kind of polymer or glass or protein or the cells won't grow. If they actually didn't know this, that's surprising and Not Cool.

More quotes:
Q: “What causes the [25%] downtime?”
A: “It really depends. Sometimes, it's contamination. Sometimes, it's a problem with the tubing."

This is a common problem with microfluidic devices (though BLI might be worse than average). Clogs, contaminants, etc.

Another customer: "It's got a lot of fluidics, and high-end FACS machines have the same problem. If you don't know how to take care of them, they become a very expensive doorstop very fast."

(Shameless plug: Nanotronics inspection solves this!)

"Cells tend to form crumbs, tend to aggregate, especially when the area, the volume needed—within the chip, it's relatively small—so it is understandable why you tend to encounter troubles as such.”

That suggests a surface chemistry or roughness issue. The trickiest thing in microfluidics is getting cells to *flow*. If they're using super-non-standard surfaces like bare silicon, maybe they just...didn't optimize this properly?

(Shameless plug #2: Nanotronics measures surface roughness at scales relevant to cell adhesion!)

Novartis manager, on cell and gene therapy:

"some of these allogeneic products would be like a 10 to 15-day manufacturing process, and the current market yield is on 65% to 70%, meaning 30% of the cells do not make it to harvest."

This isn't relevant to BLI's performance, but it's interesting as a data point that a.) cell therapy yields are crap, and b.) pharma companies are willing to spend money and try out new tech to improve those yields, even though the current market for cell therapies is small.

Of course, BLI didn't fix the yields, and the Novartis manager regretted the purchase. But the opportunity is still there if someone can get it right.

AbbVie tool user:
"What is bothering me is that I have to manually process all the images. That's a major headache and the timeconsuming part…It's tedious."

This is another WTF moment.

A Bay Area VC-funded tech company, with access to *much* better programmers than you can get in any other context, and which sells itself as incorporating AI, is making users do manual image processing?!?

AbbVie user: "Roughly, we lost 40% to 50% of the positive B cells.”

Your cells are dying! your cells are not happy in that chip! that is a lot of cell loss!

From a vendor: "The feedback was the Berkeley Lights workflows that they have defined for that particular customer are all out of the box.
There are no configurations or customizations that could be done, which is then forcing the customer to follow the Berkeley Lights ways."

WTF?
You have a new, experimental, expensive tool being sold to giant companies and you provide *zero customization*?

It's one tool designed for a variety of hugely different applications! No customization? How...how would that even work?

A LOT of these interview subjects are referring to other single-cell analysis companies (Mission Bio, IsoPlexis, and especially 10x Genomics) as strictly better than Berkeley Lights.

on contamination: "There are open areas on the media, which they tried some solutions to try to
keep covered."

there are WHAT NOW?

open, as in, exposed to the air? while you are culturing cells? and leaving them open for weeks or months?

UM.

ex-BLI employee: "It can take up to a couple of minutes to process one cell. So, that gives you a massive throughput limitation.
Just to screen one chip and to take a couple hundred cells off of that, that's an all-day process."

Some takeaways that are relevant for other companies hoping to sell to the pharma industry:

1.) There is not a ton of money in tools for cell line development. It's only $200,000 to hire a CRO to develop a cell line for you, end to end.

2.) Single-cell analysis tools and microfluidics more generally are highly competitive and rapidly advancing fields.

This stuff is getting cheaper and better all the time. It's a great time to be a *user* of these tools, but *producers* can't rest on their laurels.

3.) Pharma scientists look to academics to gauge the credibility of new tools. If academic labs can't afford your tool at all, that's a marketing handicap.