Breakfast Bytes Blogs

Today's post is about the life-science and EDA panel that took place at the recent Design Automation Conference. For my daily takes written during the conference, see my posts:

• DAC 2022: Day 1
• DAC 2022: Day 2
• DAC 2022: Day 3

I originally wrote this post and planned it to run last Monday. But, irony of ironies, it got bumped to this week because I got told about our acquisition of OpenEye Scientific, which we announced that day. The post I wrote then was Cadence Expands into Molecular Simulation with Acquisition of OpenEye Scientific. So already, Cadence is in part of the life-sciences sector.

At the end of Anirudh's keynote on day 2 of DAC, he showed the following slide:

Some of those topics on the right-hand side of the slide are adjacent to EDA...but not very adjacent. But it turns out that life sciences and biology are more adjacent than you might think. There is already work going on using EDA algorithms in biology, medicine, and brain science. On day 3 of DAC, there was a panel session on just this topic, titled What Is the Role of the EDA Community in Future Life Science Breakthroughs? The quantity of the audience was not large, but the quality was high (me for one!), with several members of the DAC executive committee and a couple of VPs of R&D from companies you know well. I won't name any names since I didn't ask permission.

The Panel

The panel was (from left to right in the photo):

• Moderator Apurva Kalia. A few years ago, he made the switch to life science after 30 years working at Cadence and is now a Ph.D. student at Tufts University.
• Lou Scheffer of Howard Hughes Medical Institute (HHMI). Lou used to be a fellow at Cadence in the digital implementation group before he switched from studying silicon brains to studying real ones. He gave a short talk at DAC 2015, which I covered in my post DAC News, Tuesday.
• Kate Adamala, Assistant Professor at University of Minnesota. Also co-founder of international synthetic life engineering initiative Build-a-Cell.
• Sameer Sonkusale, Tufts University, where he is a professor of electrical and computer engineering with appointments in biomedical engineering, and chemical and biological engineering.
• Rob Aitken. For many years, Rob was at Arm but recently made the switch to Synopsys, where he heads the Office of Technology Strategy. He was also the General Chair of DAC in 2019.
• Iris Bahar. Not on the panel but in the audience. Organizer of the panel. She is at the Colorado School of Mines, where she heads the CS department.

Apurva opened with a couple of slides making a direct comparison between EDA and life sciences:

He used these to introduce the two questions he wanted the panel to discuss:

• Are there knowledge/methods/techniques that can be transferred from EDA into the world of bio/life sciences?
• How easy/difficult is it to enter the field of bio/life sciences?

Introductions

Each of the panelists then had a few minutes to give a summary of their position.

Lou went first and pointed out that EDA already contributes to life sciences.

One example is Rent's rule (which relates the number of I/Os required for a block containing a given number of computing elements). In both cases, EDA and biology, 0.5 to 0.7th power of the number of "gates" gives the number of "pins." He then listed several papers that connect EDA algorithms to life science.

Next up was Kate Adamala.

She pointed out that nature did not sample all the possible ways to build life. It is really an n=1 situation. Note the image source for the above slide, Charles Darwin.

Biology is pretty boring from a chemical point of view. We all use the same ways to build proteins and so on

For her, the ultimate challenge is synthetic cells, new cells with new architectures built out of truly interchangeable parts (IP!) in a different chassis.

Sameer came next.

He had a vision of truly personalized medicine, such as healing wounds with smart bandages and how to overcome the fact that brittle silicon is not a great substrate for putting in the human body. He talked a little about bioinstrumentation, such as analyzing the gut biome not by analyzing *** but by swallowing a pill and tracking it as it analyzes the gastrointestinal tract as it passes through.

Finally, it was Rob's turn.

He started off doing an inventory of what EDA is good at (see the slide above)

• Adaptation
  • Use available compute power to boost capacity, performance
  • Compiling circuits into machine code for simulation
  • Map circuits into seas of FPGAs for emulation [or a specialized ASIC in Cadence's case]
• Simplification
  • Restrict design to enable abstraction
  • Clocked feedback-free digital logic
  • Confine analog behavior

Discussion

As always, when I try and write down a panel session like this, I precede the questions with "Q," so they stand out. That means they were asked by Apurva. My annotations are in [brackets], meaning that is not something said by the panelists. And this is not a transcript. It is more like how network TV broadcasts the Olympics when it is in a bad timezone as "plausibly live." When questions were asked from the audience, I precede them by "Q Audience."

Q: In life sciences, the physics is not well understood. It works well in EDA because the physics is understood. So what kind of modeling can we do?

Lou: We can use modeling techniques backwards. We have neurons, we know the synapses, but we don't know the strengths. So we can model it and see if we can get the behavior we observe.

Kate: We run into the problem of the physics not being understood all the time. We still don’t know all the components of the cells. It is easier to build bottom-up and see if one day they start to behave like a cell. From there, we can build predictive models. That might or might not work. We haven’t done it yet.

Sameer: Models don’t exist because engineers have not really attempted to build them yet.

Rob: Look at how EDA was able to be successful. It was the restriction piece. If you can just arbitrarily combine transistors, you end up with a mess. But if you restrict them to CMOS and then build gates and clocks, you get something tractable. We need to pick and abstraction level. You can’t do arbitrary things in chips, but those restrictions don’t exist in biology.

Q Audience: What is the equivalent of the standard cell? The models may not be complete, but is there an equivalent?

Kate: We don’t even have a satisfactory definition of life, which is a big problem.

Rob: If a biological standard cell existed and you could do things with it, we could handle abstraction.

Lou: Maybe the reason life succeeds is that it has lots of systems that don’t interact.

Q Audience: It looks like EDA + Life Science is a small field [gesturing to the small audience]. Variability, system complexity, reliability. It looks like these are all really difficult for bio-life. So how do we tackle them?

Lou: Biology can probably help. Look at circuits behind a fly’s eyes. They are very different. Biology works despite huge amounts of variation, something we might need to learn how to do.

Iris: We are dealing with this in the EDA world. We dealt with it many years ago, building reliable computing systems from unreliable components back in the 1940s, and we are dealing with it again now.

Lou: Biological systems are amazingly resilient. If you knock out any link, it still works, but maybe not quite as well. If 50,000 gates died in your phone, you’d notice immediately.

Q: How do you model biological systems?

Kate: Closest is where we watch the progress of a single molecule like ATP [adenosine triphosphate, a key part of Kreb's cycle that delivers all the energy biological cells use]. That’s following molecules rather than following structures. These produce the best results. But when you model neurons, you don’t care at that level.

Lou: We can do some high-level modeling. In a synapse, we just treat mitochondria as generating ATP.

Rob: Think about how EDA would have evolved if, in 1940, somehow a modern state-of-the-art supercomputer suddenly appeared, and everyone spent decades working out how to build one. Life science is more like that. You have a 3nm chip, and you can’t even see the wires because you need equipment that has not been invented yet.

Lou: Synapses we treat as just working, like gate level. Then we have organ level, more like IP. But we need some new models for the biological world. There are 100 chemical products in a synapse, but we can model it simply. Start with SPICE and then move into custom languages for the domain that we are examining. At the high-level we don’t have anything, just Python and MATLAB.

Q: How will it be different if we succeed in this collaboration?

Lou: Sameer mentioned intelligence in band-aids, but there is no need to restrict them to people who are stuck. Blow your nose, and it tells you what kind of cold you have, or analysis in your toilet. There are far too many stories of people discovered with cancer just because they were X-rayed for a broken wrist or something.

Kate: Safety and security. One day we will make biology more efficient, but then it becomes a tool that can be misused. We need to maintain not just our current standard of living but keep ourselves alive through the next pandemic or whatever. We just need to design biology to be better.

Sameer: I can see direct impact with designers engaging with life-science and biology, making sensors, and diagnostic tools. More and more people are getting into the field. We will have better models of understanding biology. Issues we haven’t thought of yet.

Rob: There are many more nightmare scenarios in life-science than electronics, and we’ve not done a great job of security there.

Q Audience: It took billions of years to come up with the [biological] cell. Then we had a Cambrian explosion with lots of creatures.

Kate: Once we have a synthetic cell, we can build up complexity in ways we cannot predict for now.

Lou: There are lots of chances for this. Every single generation has to work, so you can’t make big changes. If you could design creatures from scratch could build better, like airplanes are faster than birds.

Kate: Everything is made from proteins, and ribosomes are not even very good and are terribly slow.

Rob: Over time, transistors have got worse since, as you shrink them, they need more power, so a ribosome may be a really bad way of doing it, but perhaps it’s the worst way except for all the other ways.

Lou: We don’t know how to build enzymes from scratch. We just copy existing enzymes. So there is lots of room for improvement.

Q: How do we do the validation step of any synthetic biology?

Lou: There is one big validation step: can you produce children?

Kate: If I’m building a strain that makes ethanol, that's the validation step. Biology makes things that make more things.

Sameer: We can have a kill switch to turn it off, but can you turn it on again. Creatures cannot reboot.

Q: In EDA, we have CS, math, engineering, and so on. In life science, biology and chemistry. How do we get students to be interested in this? Excited about it?

Lou: A lot of engineering schools do a sort of survey course in the first year. It seems it should have something like that “life science for engineers” like those "physics for poets" classes. So they can get a feel for whether they are interested. But engineering schools don’t do much in bioscience if any.

Q: Four and a half years ago, I decided to switch. Hence going back to school and being in a Ph.D. program. My daughter, a biologist, was one of my teachers. It wasn’t that difficult to switch. Every day is something fascinating. So tell students when they come in about how it can happen. Students don’t realize switching can even happen.

Sameer: We need to solve this problem, or we all die long-term.

Kate: It would be great if biological students could get exposed to engineering. I have people who haven’t a clue about Python. “Wow, you know how to code," I get told.

Rob: Cross-disciplinary activities are really important, even if you never use them.

Q: Well, we're out of time, so let's thank the panel.

 

Sign up for Sunday Brunch, the weekly Breakfast Bytes email.

.