SAS and Open-Source – How They Integrate and Other News From SAS

Transcript

SAS and Open-Source – How They Integrate and Other News From SAS

[00:00:00] Alexander: Welcome to another episode of the Effective Statistician, and today I’m super excited to have someone from a very, very big [00:00:10] company that I first got introduced to something like 30 years ago, so a very, very long time, and that is [00:00:20] SAS. Hi Mark, how are you doing?

[00:00:22] Mark: I’m doing well, thank you Alexander, thanks for having me.

[00:00:25] Alexander: Very good. Yeah, I still remember my days very early in [00:00:30] university when I got my first statistics courses, and I was confronted with SARS. I think it was kind of version 3. 1 or [00:00:40] something like that at the time. So from that, you probably can tell how old I am. And so I had these kind of books for SAS, for [00:00:50] statistics, for beginners. And I can still remember working with that and doing my first data step and these kind of things. 

[00:00:59] Mark: [00:01:00] I’m with you there. I’m with you there. 

[00:01:01] Alexander: Yeah, yeah. Okay. So but this is not about me. This is about you. Tell us a little bit about yourself and [00:01:10] how you got to where you are now? 

[00:01:12] Mark: Absolutely. Alexander, my name is Mark Lambrecht. I’m heading up health and life sciences team at SAS, part of the global leadership. [00:01:20] I’m years at SAS this year. Started in 2005. I’m actually a Belgian. I live in Leuven in Belgium. And I’ve [00:01:30] started, I’m educated as a bioscience engineer. I’m a specialist in cell and gene genetic engineering. And about yeah, 25 years ago, you might remember [00:01:40] that there was a lot of news about genome sequencing and how that got to.

[00:01:45] Mark: Start that kind of designs and that’s, that is where I was in the lab, but I [00:01:50] rolled into bioinformatics trained as an engineer not as a statistician as yourself, but as an engineer, we like to tackle different problems. And the problem we had at the time was. [00:02:00] Large amounts of data, genomics data and statistics not being used traditionally in that way.

[00:02:07] Mark: Looking more at how engineers were using early [00:02:10] machine learning, data mining and combining these techniques. And so I started applying that to to the field. I then went to a biopharma company called Galapagos [00:02:20] after that. Worked as a bioinformaticist and found out that SAS has a vibrant life sciences and healthcare customer base and we’re very active in that [00:02:30] domain.

[00:02:30] Mark: So I stuck with that and started in Belgium and expanded internationally and still here, still working for life sciences and healthcare company together with many [00:02:40] colleagues with the community out there focused on, on the different things we do out there, which is statistics, but a lot of things outside of that as well.

[00:02:48] Alexander: Awesome. Awesome.[00:02:50] I for a very, very long time only worked with SAS. Well, maybe with some things like and query and other things, [00:03:00] but, but mainly with SAS and SAS was the go to, and it’s probably still the go to tool for, for many of the things that we do within the [00:03:10] pharmaceutical industry. Over the last years other open source packages, especially in within the pharma industry. I think it’s a lot of around [00:03:20] are have evolved. And recently I learned that source is embracing this change as well and integrating [00:03:30] these kind of languages into the overall sales landscape. How does that work? 

[00:03:36] Mark: Yeah, thank you for that, that question. And I appreciate the [00:03:40] opportunity to explain that clearly most people know SAS from the language, right? From using the prox, the data steps that we have. But SAS is [00:03:50] much more than a language. Already 20 years ago, we started creating technology to manage data industry solutions to help with clinical trial data [00:04:00] or in the healthcare hospital area, applying data mining machine learning techniques, AI techniques.

[00:04:06] Mark: So I would say that since as [00:04:10] long as I’ve joined SAS. We have been using open source access to create those solutions, for example. And if you look at what is happening in the technology [00:04:20] sector, you’ll see that every technology company out there is using open source to strengthen what they do.

[00:04:25] Mark: They built on top of open source frameworks. They built solutions on top of [00:04:30] that. But you’re asking specifically about the language and R and maybe other languages like Python to some extent for machine learning or maybe less prevalent languages, [00:04:40] but that will become more prevalent like Julia.

[00:04:42] Mark: Interesting as well for some techniques and we have completely embraced those. So it’s not just no longer about the SAS language. What we want to do [00:04:50] is bring the right analytical. Technology in a very specific way in a validated way to our users. And so all of our solutions are currently [00:05:00] using not just the SAS language, but can also allow users to program in our, for example, if you think about our clinical trial solutions.

[00:05:09] Mark: And [00:05:10] yeah, I think. That that is really important to understand that that we have made that switch that we’re embracing that what we want to bring, however, is focus [00:05:20] on the right analytical algorithm, and I think there’s a lot of value to do that. And sometimes we have created that ourselves and we have hundreds of PhDs in statistics.

[00:05:29] Mark: [00:05:30] Everybody knows, I think, what the quality is of what we bring out there, the documentation and the whole framework around it. But if a technique is not available in SAS, people need to be able [00:05:40] to go to R and program in R or in Julia or in Python, find it and mix and match it into that analytical framework in an end to end way.

[00:05:49] Mark: [00:05:50] So yeah, there’s many other areas as well, because if you talk open source, it’s not just about the language. And probably we’ll talk a little bit more about that as we go, Alexander, but [00:06:00] it’s also about the interoperability and that’s also where we. Heavily invested in in the last few years, the ability to open up what we bring with, for example, REST [00:06:10] APIs or building blocks that you could call from R and that you could call those routines that you stay in an R or in a Python or in any other language environment.[00:06:20] 

[00:06:20] Mark: And you call the SAS data step and the procs that you know from many years ago, and you just call them in. And you have access to that technology vice versa. If people that are still [00:06:30] in SAS language and a lot of pharma companies need that, I can also call in the, our language routines and the packages that they need to, to bring [00:06:40] in the data and switch data results back and forward as they need it in that hybrid and best of breed environments.

[00:06:47] Alexander: How does that [00:06:50] practically work? Is it some kind of proc R and then you work from there or how do you kind of [00:07:00] call R from a SAS environment? 

[00:07:03] Mark: Yeah, there’s different techniques currently. There is indeed the ability in R to call a SAS routine [00:07:10] specifically using what we call a SWAT package. That is a package that exposes some of the SAS capabilities so that you can natively call it [00:07:20] into R. The same exists for Python. There is also the ability to in SAS, for example run the R capabilities with things like IML.[00:07:30] So interactive matrix language and call in R and even call the R code within your SAS environment as well.

[00:07:37] Mark: If you think about Python [00:07:40] specifically, our editor, which is called SAS Studio, you can program in Python and we will likely open it up to program in R straight from that editor, but you can remain to work in a [00:07:50] Jupyter notebook or in an R studio. Environment and also call the SAS routines in there with that package specifically.

[00:07:57] Alexander: There’s also both ways?

[00:07:59] Mark: [00:08:00] Both ways, absolutely. And the REST APIs R and Python and other languages have capabilities to call on REST APIs. And so those end points are opened up because we [00:08:10] basically moved all of our technology stack in a cloud environment in an open source. Kubernetes environment so that you could call it in there as well.

[00:08:17] Mark: And that is microservices enabled. And that [00:08:20] allows us that latest generation technology allows us to open it up from that point of view as well, so that you’re not only calling the analytical algorithms, but [00:08:30] also the data that is stored in a secure way that you would have one environment to do that.

[00:08:34] Mark: And usually as you talk about analytics quickly, you go beyond just [00:08:40] running the analytical routines and think about, okay, how do I store my data? How do I. How do I hand it off to other systems in this world of clinical trials? And so [00:08:50] all of that, could I call them enterprise capabilities are needed and that’s where we come in and provide that capability.

[00:08:57] Alexander: Okay. So, and so [00:09:00] basically you could have all your, you know, existing SAS environment versus our data sets, all these kinds of different things. And could pull, [00:09:10] then pull from. From our, these kinds of different things and storage and back into these SAS environments. 

[00:09:18] Mark: You could do that. [00:09:20] We see a lot of movement in the cloud area in terms of how data is stored. So yes, SAS tables are still essential, very much like our data frames are needed because that’s [00:09:30] how a statistician thinks in a tabular format, but clearly if you think about. complexity of data these days. It’s not just about tabular data. It’s unstructured information. It’s [00:09:40] about types of information, and that could be stored in some cloud services.

[00:09:44] Mark: If you think about Azure data services or AWS S3 or [00:09:50] those type of buckets, and you could call them in either in R or in SAS and transparently move those over as well in a one secured model. So [00:10:00] Unlike people are thinking about SAS and SAS tables, yes, that is still important and we will keep investing in that, but we have different data formats and we can interact with parquet data [00:10:10] formats as well as you look at larger amounts of data as well in that new technology.

[00:10:15] Mark: So I’d like people to think about SAS as having evolved [00:10:20] from that one language that was the SAS 9 environment and compiler and PC SAS to a modern cloud environment where you can mix and match technologies. If [00:10:30] people want to use R, we will welcome that. We see particular strengths of using SAS, by the way, which are not to be found in R.

[00:10:36] Mark: My position is that we will end up in a hybrid [00:10:40] fashion in where people will use the technology and the language that they feel most comfortable in. But if you think about statistical techniques, you will still need to have well documented and validated techniques [00:10:50] and certainly in the pharma industry.

[00:10:52] Mark: And that could come from either way. There could be R packages. It could be SAS capabilities. There could be other capabilities like coming from newer [00:11:00] types of engines as well, especially as we think about generative AI and those type of techniques. So it’s really a new world out there also for SAS. And I invite people to [00:11:10] have a look at that.

[00:11:11] Alexander: So in terms of all the R packages that come out on a probably kind of daily basis or even [00:11:20] faster. How do you bring them into your source environment? How do you kind of qualify them? 

[00:11:27] Mark: Well, there’s different ways to do that. [00:11:30] I can’t say there’s one single way in we dictate how a user or an organization needs to use that. I see, like you say, it’s a bit of a Cambrian [00:11:40] explosion, probably of ways to deal with that. And there’s probably also trains of thought on how to do that. And I see a [00:11:50] lot of probably 95, if not more percent of companies using qualified environments for clinical trial analysis that are very much [00:12:00] leaning towards SAS with some R involved in that.

[00:12:02] Mark: And you see companies moving into kind of those mixed environments or setting up an R environment. And then the question [00:12:10] comes, how will we manage that? And what we do is we host that R environment. And we, we lock it down. for the users. Now, if users want more [00:12:20] flexibility in terms of package management and how they do that, we leave it up to them.

[00:12:25] Mark: And we listened to the organization, how they’d like to do that. And I see different pharma companies having different[00:12:30] ways of doing it. Some pharma companies really want to manage kind of all different combinations of package versions and invest a lot of time and effort and people in it. But most [00:12:40] pharma companies say, okay, at some point we want to have a look at that.

[00:12:43] Mark: But we want to ensure that it’s stable at the level of when we use it in a clinical trial. And that’s where [00:12:50] we can come in and help them manage that stability and that that kind of qualification behind it. But the bottom line is we leave it to the organization to decide how they want to do [00:13:00] it because there’s different styles of how to do it. And I don’t think the industry has settled on one specific way. 

[00:13:06] Alexander: You just mentioned generative AI. Yeah, [00:13:10] so everybody is talking about chat GPT and all the other resources there. And I’ve learned that a lot of people kind of use [00:13:20] it to create code and that it’s actually quite, quite good at it. How does that work? Do you already have some kind of [00:13:30] generative AI solution within SAS? 

[00:13:32] Mark: We do and we announced that our user conference a couple of months ago in the major SAS user conference in the US and Las Vegas, we mentioned [00:13:40] the availability of a co pilot like functionality. And so we have a partnership with Microsoft in which we have early access to the open AI.

[00:13:49] Mark: [00:13:50] Libraries. And so what we have done, obviously we have a lot of SASS codes lying about, we decided that we want to bring out, out Gen AI or LLM like [00:14:00] functionalities that make sense for people that differentiate what is already out there. And for us, that is having sasco. And so we are training with those RAG the [00:14:10] retrieval augmented generation techniques.

[00:14:12] Mark: In addition to the foundation models. We’re training that on sasco and so that people can. We will be able in Visual Studio Code, which is yet [00:14:20] another open source effort that is very valuable. We have created a plugin for people to use SAS editing capabilities that you will be helped by that copilot like functionality.

[00:14:29] Mark: And that will be [00:14:30] coming out sometime this quarter in, in a first production release. In addition to that, we’re looking also at The different types of use cases, as [00:14:40] you can imagine, and we’re already far ahead with some pharmaceutical companies in in driving those capabilities, looking at unstructured text information.

[00:14:48] Mark: The idea is that we will end up with [00:14:50] intelligent assistance that allow people to take unstructured information that really needs to remain private. And as you know, A lot of pharma companies have a lot of information [00:15:00] that they don’t want necessarily to throw out in the public before having analyzed it.

[00:15:03] Mark: And so those workflows are currently available, I would say many in our [00:15:10] latest generation SaaS via platform that allows you to, to embed that, to use text analytics, combine that with LLMs. We don’t only do that with open [00:15:20] ai, by the way we do it with open source since the, the theme. Of, of this podcast is, is also open source and SAS.

[00:15:26] Mark: We do it with open source and we take like Lama models and [00:15:30] look at that, how, how performant that is. So there’s a lot of I would say piloting a lot of use cases going on. Some of them already being used. But you will see [00:15:40] more of it appearing in the Visual Studio Code extension later on also in our programming environments.

[00:15:46] Mark: What it does for us, it allows people to program in SAS [00:15:50] faster, but also I see a lot of help in all languages. It will, in my prediction, be less important in what language you’re programming. So the whole debate [00:16:00] Alexander of R and SAS and, and where to go and languages. I think we really need to go beyond that and think about what gen AI will do.

[00:16:08] Mark: And I think it will be easier to [00:16:10] move from one language to another in, in in all sorts of directions. And maybe that’s what we need. We need people to speak in, in the analytical language and look at data in a [00:16:20] meaningful way where you can kind of have chatbots helping you generate the code that you need to interact with that.

[00:16:26] Mark: Yeah. Engine. And of course we want to be that engine at SAS. We [00:16:30] want to remain that qualified engine that gives you the right statistical algorithms because we all know that Gen AI also is hallucinating at the moment[00:16:40] and no statistician will appreciate that of course. And so the ability to put guardrails around that is what we’re focused on as well.

[00:16:49] Mark: Yeah. 

[00:16:49] Alexander: [00:16:50] Awesome. Awesome. Yeah. Yeah. I think this is a. Yeah. Whole new world coming out. The other point is, I really love that you can come in [00:17:00] with all these different experiences and skills and brings us together because We have so many people that [00:17:10] are 100 percent SaaS programmers, and then we have a lot of young people coming from the universities that never have touched SAS. And [00:17:20] so helping these to work effectively together in teams. Is, is really, really important. 

[00:17:27] Mark: I agree with that. Yeah, I agree with that. And there’s a lot of emotion [00:17:30] in that debate often. And I think we need to, I see that too. And I saw that at FUSE, my message at the FUSE open source discussion panel in [00:17:40] Birmingham.

[00:17:40] Mark: Couple of just a couple of months ago was let’s rise above that debate and make sure that indeed those that come from university and know one language that they learn [00:17:50] other approaches and other languages. That’s what I have done. When I started programming, I used to do Pearl programming, which was an open source language.

[00:17:57] Mark: Very good for bioinformatics now. Probably a lot of [00:18:00] people need to look up what that means. Today it’s Python. But it’s no use sticking to one approach. By learning a new language or a new approach, it learns you to think about data in a different [00:18:10] way. I think I’m convinced of that. And that’s, that’s really where we need to end up.

[00:18:14] Mark: I think that’s what pharma companies need to do, right? They need to think about patient data and make sure that they [00:18:20] have seen every aspect of it in terms of efficacy and safety of the, of the drug. And there’s just all these new, exciting technology out there that, that can help us to do that. So that, that’s [00:18:30] great.

[00:18:30] Alexander: Yeah. Yeah. And there’s a lot of companies in investing in that area. One of the things that I always[00:18:40] Loved R4 was were all the data visualization capabilities with ggplot and, and these kinds of things. And this lots of [00:18:50] people said it’s, it’s super easy that way. And. When that is now easily integrated into, into SaaS that makes [00:19:00] makes a combination super, super helpful.

[00:19:03] Alexander: Because yeah, getting the statistics, the, all the algorithms from SaaS and then combining [00:19:10] with, with the Kremov graphics from R makes a very, very beautiful combination. 

[00:19:16] Mark: Yeah, that’s absolutely true. And all of these visualization [00:19:20] techniques and I will give that R and R Shiny have opened that up, right?

[00:19:23] Mark: They have opened up the, the vision of people to be able to create interactive applications.[00:19:30] I would say there’s also beyond that different techniques to, to scale that up even we have, we embedding it in an enterprise fashion. [00:19:40] But also thinking about visualization dashboards making data available more rapidly in an interactive way and still have the backend code available to, to look at [00:19:50] what you exactly did at, at what point even going as far as building apps and having apps that if you have a machine learning model, or if you have some model where it’s programmed [00:20:00] in RSS, Python, whatever.

[00:20:01] Mark: That you can embed that in the so called app factory, which we were, we have announced as well. And so you see these techniques as [00:20:10] well, what people need to think about is scalability reuse. I think that’s often a little bit underestimated the effort going into one app versus doing it together in a [00:20:20] collaborative way and respecting your it colleagues who are asking you to adhere to enterprise technologies.

[00:20:26] Mark: And that’s where where we want to add value as well. And look at that and [00:20:30] scale it up if you want. 

[00:20:31] Alexander: Yeah, because that’s where really see the value comes in. Yeah. When we have these [00:20:40] interactive data visualizations that we can reuse again and again for, let’s say for. For safety, yeah, you will need to [00:20:50] look into lots of lots of studies on an ongoing basis and check that everything is okay from a safety perspective.

[00:20:57] Alexander: For every database look, you want to [00:21:00] have something like an interactive data visualization that you can very easily understand. What is all the data about if you can [00:21:10] quickly answer what kind of different questions about subgroups, about things like that. I know that there’s some big pharma companies that invest a lot in [00:21:20] this and make sure that they have these kind of interactive readouts after every database log.

[00:21:27] Alexander: And that is hugely beneficial [00:21:30] from upper management. Bye. Just talk to someone at one company that is currently a little bit under a resource pressure. And[00:21:40] he mentioned that, well, lots of things get cut, but not this interactive data visualization readouts because it is so, so helpful. And so, and that [00:21:50] is of course, where exactly. These two things come together. Yeah. See the power of good algorithms, a validated environment, interactive [00:22:00] graphics. And you need that again and again and again. 

[00:22:03] Mark: Yeah, that’s absolutely true. And I would say it goes even beyond the clinical trial reporting. If you look at data [00:22:10] quality and that’s how that is a responsibility for everyone.

[00:22:13] Mark: In the clinical trial, it’s also useful for clinical data management. If you look at where data is now coming from [00:22:20] in terms of sources, more from decentralized sources and electronic patients, all those types of systems and having the [00:22:30] ability while the data even is coming in to visualize that type of data with enterprise functionalities and rapidly see what is happening to your data is essential.

[00:22:39] Mark: And that’s [00:22:40] how we look at it as well. And there will be no surprise that that we have built it up enormously. And I invite people also to look at that and see what they can do with their current [00:22:50] skill sets, because there is the ability to work together here with different profiles as well. And that’s also a benefit I think of the open source mentality is, is [00:23:00] collaboration, right?

[00:23:00] Mark: So that’s, that’s something we want to increase as well as people yeah, gather their technologies and their talents and how you can really make that available. [00:23:10] Not think about it in your own way. I create my own app visualization app and I hand it off, but how does it work for others and how can I scale it up? So that’s really important. [00:23:20] 

[00:23:20] Alexander: Yeah, yeah. One of the things that I keep talking about is instead of creating something new again and again and again, [00:23:30] maybe just think about how you can share what you have already created more and more so that more people can benefit from it. Since that is [00:23:40] a completely different value statement.

[00:23:43] Alexander: 1 thing is truly important here is sees a volume of data. As you just [00:23:50] mentioned beyond clinical trial data data that comes from medical devices or kind of different things. If you think about patients with diabetes, they have kind of, [00:24:00] you know, monitors that 24 7 measure all kinds of different things.

[00:24:05] Alexander: I’m pretty sure this is just the start of lots of [00:24:10] different things. If you think about movement disorders, if you think about eye disorders, if you think about skin disorders, there are so many, you know, [00:24:20] applications where there will be kind of continuous A stream of information coming in where people will get, you know, monitor their [00:24:30] disease on an ongoing basis.

[00:24:31] Alexander: And if that affects millions of patients, then well, you’ll get a lot of data coming in [00:24:40] and, and managing that in a, in a useful way will be really, really important. Last question I have is when people go to [00:24:50] SARS, what are kind of useful references for people to have a look into, so that they learn more about the open source integrations with SAS?[00:25:00] 

[00:25:00] Mark: Yeah, there’s a couple of resources and possibilities available and by the way, you just mentioned the large amounts of data and I can’t help but point that some of the healthcare [00:25:10] work we have done at hospitals that shows how important it is to do prediction of risks on patients and in which we are.

[00:25:16] Mark: Use even streaming analytics analytics based on data [00:25:20] that is in movement and in which you can do that. That’s not yet done a lot on clinical trials because these volumes are still different and those biomarkers need to be validated, but you see it in real world [00:25:30] data. You see it in other areas.

[00:25:31] Mark: As to your question, Alexander, on where do you go? Well, we have different resources. We have a communities. sas. com [00:25:40] environment where there’s a lot of information being shared on the latest of what SAS is doing in that area. We have a developer environment where people can look at our REST [00:25:50] APIs and look at the libraries and open source collaborations that we have.

[00:25:53] Mark: That’s on developer. sas. com. We have a GitHub available. We publish a lot of codes in different [00:26:00] varieties and for different projects on our github. com forward slash SAS software URL if you want to go there. And I would also invite you just to look at the [00:26:10] conferences that we present at, at FUSE, at CDISC, at where we.

[00:26:14] Mark: The way they can find a lot of information on that, and you’ll find information on Visual Studio code, you’ll find information on the [00:26:20] REST APIs and on the open source integration, also on the AI machine learning front where we do model management and things like that. So documentation, of course, is [00:26:30] one other area.

[00:26:30] Mark: Our our famous documentation that you can go to and where you can find also the latest techniques that we’re implementing in different AI machine learning techniques as well as we [00:26:40] have evolved that story enormously, but I’m also happy to, to be taking emails or response to specific questions.

[00:26:47] Mark: And my team is ready as well. We like to [00:26:50] also work together with people if there’s we’re working together with projects like data sets, Jason. And if you see this effort, we’re working we were using now CD score as efforts and we will [00:27:00] be publishing on that. And so if you go to the papers on Lexie Hanson, for example, dot com, you’ll see a lot of papers that we have published in different conferences as well as it’s a, an ever [00:27:10] evolving field.

[00:27:11] Alexander: Awesome. Thanks so much. And if you couldn’t kind of write down all these kind of different links for the moment. Just head over [00:27:20] to the corresponding blog of this episode at theeffectivestatistician. com and you’ll find all the links there and also a link [00:27:30] to Mark. 

[00:27:32] Alexander: Very good. Thanks so much. We talked about your journey from getting into SAS and how [00:27:40] SAS is, you know, developing from a kind of Stand alone a system into a really big enterprise systems that [00:27:50] has lots of different interfaces with open source, but also with and other areas, how AI generative [00:28:00] AI, all these kinds of things come into play.

[00:28:03] Alexander: So visualization can work interactive visualization, how we can manage really [00:28:10] big data, so to say, and the healthcare industry. And that was a wonderful, wonderful interview. Thanks so much for being on the show, Mark. [00:28:20] 

[00:28:20] Mark: Thank you, Alexander, for the opportunity and congrats with the fantastic podcast series you’re running here.