Please note transcript has been edited for brevity and clarity.
FLG: Hi everybody. Ahead of The Festival of Genomics, Biodata & AI in Boston this summer, I'm speaking to a few of our expert speakers. Today, I'm with Anas Al Zabiby from Novartis. How are you doing today?
Anas: I'm doing fine. How about yourself?
FLG: I'm doing well. Thank you. Now, I think the best place to start is to tell the audience a little bit about yourself and your background.
Anas: I'm a PharmD by education and I switched to bioinformatics about eight years ago. I joined Novartis around the same time as a data scientist. I worked in discovery and development, I worked on clinical trials data, and I worked in the multi-omic space. Right now, I'm part of discovery sciences in Novartis Biomedical Research, and I work on multi-omics integration and AI, along with also working on human genetics and biobanks.
FLG: Your talk at The Festival is going to focus on multi-omics, and that can offer quite a holistic view of biology. But when you're planning a same sample multi-omic study, what are the key decisions that determine success or failure?
Anas: Well, there are multiple factors, but if we want to generalise, I would say the study design itself. If you have a study design that accounts for the biases between technical and biological, that's always key for deconvoluting and understanding how the data, when it's ingested into an AI algorithm, is being interpreted by it, or understanding what you expect the output to be.
Technology and the types of omics that you're using play a role, and the harmonisation of data and processing. If your data is not harmonised well, it might not scale or produce the outcome that you'd expect it to produce.
And finally, the computational methods you have employed [play a role]. There are so many different ways of analysing data. In the context of multi-omics, you're talking about different ways of being able to analyse these different omics, and based on the technology, the sparsity, and the variability of different omics, you would choose the right methods to do that. Of course, following harmonisation and processing, etc., considering the design of your study and, let's say, minimising or mitigating bias to the best degree possible. I mean, if you think about it, you have multiple layers. You can either analyse them separately and then try to compare them at the end, or you can combine them from the very beginning and analyse them from the very beginning, together as a single piece, and then try to draw conclusions from all of them together. So, there are multiple ways. There are different levels of complexity to the methods as well, ranging from simple statistical methods to machine learning to deep learning methods, like graph convolution nets.
FLG: Another aspect of your talk is that you're going to be focusing on AI and ML, and there's a lot of hype around this in research at the moment. But where is it actually helping in 2026, versus what's still experimental?
Anas: The AI hype is coming from the LLMs, large language models, ChatGPT. There's a lot of hype, but low success rates, and that's being talked about right now. But AI is also starting to find its place in the world, I think. It's creating faster turnarounds, we're able to do things a lot faster right now. And I think, yes, there's hype, but there's also some realistic advantages to deploying and employing AI in research. That's what we do. As a company, Novartis is big on AI and data.
FLG: In the context of multi-omics specifically, how can we effectively validate the findings from AI? And do you think people trust those findings, or is there still work to be done?
Anas: I mean, this is applicable not just to multi-omics, but to a lot of research that involves complex computations. It really comes down to certainty and confidence in the outcomes you see.
I would say there are two main ways to validate findings. The first is experimental validation. For example, in large-scale multi-omics experiments, we look at different angles of a disease. Once we have findings, we can follow them up with low-throughput experimental techniques—looking at a few genes or proteins—to understand whether the signals we observed are true and actually associated with the disease.
The second way is computational validation, such as parallel analyses or meta-analysis. If you perform an analysis on one dataset and can replicate it in another dataset in a similar way, you can be more confident in the outcome, within the bounds of the experiment.
So, you can either computationally replicate your findings in another setting, or experimentally validate them when possible. These are the best approaches, and they’re always essential.
FLG: And in your experience, have you seen that organisations have been willing to adopt these really cutting-edge AI methods? Or do people want to stick with more established, interpretable approaches?
Anas: Yeah, that’s a good point. In a way, we’re always trying to stay on the cutting edge of technology—both in computational AI and elsewhere. But at the same time, we want to establish a ground truth or a standard reference to compare the success of any new methods we’re trying.
So, when we deploy new, disruptive AI methods, we always compare them to something simpler—something more interpretable and easier for us to understand. Whenever we’re doing something complex, we try to break it down so we can better interpret it.
Either the method needs to be interpretable to some degree, or we compare it to a gold standard that’s already established. In most cases, we do one or both. Otherwise, we can’t confidently say we should adopt a new method.
That said, we are always looking to stay on the cutting edge and to explore and apply new methods.
FLG: As a scientist, what do you prioritise the most when it comes to choosing between competing tools or methods?
Anas: Outcomes. The question has to be addressed.
The aim is not about choosing the most complicated method. The goal is to answer the key scientific question. So, we typically move from simpler to more complex methods, step by step, as needed and as appropriate for the question.
We don’t try to generalise one method for all situations. Especially in multi-omics, there’s data sparsity, signal sparsity, and a wide range of study designs and technologies.
So it’s really about tailoring the method to the key scientific question, based on the data you have.
FLG: That makes a lot of sense. What are the most meaningful metrics for benchmarking in the context of your work?
Anas: That’s a good point. We usually start with the literature—systematic reviews of existing evidence and methods that have already been deployed and compared.
For example, in multi-omics, there are graph convolutional network approaches. The MOGONET paper from 2019 benchmarked on datasets like TCGA, and then later papers followed, comparing against the same datasets and methods.
We take a similar approach—comparing our methods to previous ones on the same datasets to see how they perform. Are they doing better, or not as well, on those benchmark datasets?
That said, benchmark datasets can be quite specific. For instance, the TCGA dataset used for benchmarking is a breast cancer multi-omics dataset, which may differ from the data you’re working with.
So there can be a gap, but you still benchmark computational methods and then try to bridge that gap to your own data. Overall, benchmarking is largely based on literature and the datasets used in that literature.
FLG: Where do you see this field going in the next five or ten years?
Anas: I think the future is multi-omics. It’s going to grow in scale and become more disruptive—we’ll be able to treat more diseases and develop new therapies faster. It will also help address population disparities in precision medicine, which is something we always strive for.
With more studies deploying multi-omics and new methods emerging, this will become increasingly possible. Right now, there’s been a lot of hype over the past few years, as more people start working in this space—similar to what we saw with next-generation sequencing about ten years ago. And that wasn’t wrong—it was the right direction.
Similarly, there’s now strong evidence that multi-omics works, and many people are investing in it. In my opinion, this is the right direction.
In five years, I think it will become more standard and systematic. Study designs will become clearer, along with how to deploy multi-omics, which technologies to use, and how to apply them across different diseases.
I’m excited about it and looking forward to seeing how it evolves—and hopefully contributing to that progress as well.
FLG: And something we really care about when organising The Festival is facilitating connections and collaborations between people. As we move toward a multi-omics or AI-driven future, it’s not just about building new tools—there’s also a lot of cultural change that needs to happen. Why is that so important, and do you think events like this can help people connect in that way?
Anas: To be honest, I think it naturally evolves. In places like the Cambridge area, where we work, the scientific community is always striving to advance therapies and deepen our understanding of biology.
So learning and adopting new technologies—like those in multi-omics—happens naturally. But connecting with people is still essential.
Events like The Festival of Genomics provide a great platform for that, especially with a pharma focus. It brings an industrial perspective to multi-omics, compared to academia, which is often more exploratory.
There are real opportunities there. Connecting with people—whether in sessions, chat rooms, breakout rooms, or meetings—acts as a catalyst for collaboration and helps spark new ideas.
FLG: And what are you most looking forward to about the event?
Anas: I’m looking forward to connecting with people and learning about new technologies from colleagues who will be attending, as well as the work others are doing. Most importantly, I’m excited to connect with experts in the same field.
There’s going to be a multi-omics platform, and I’m looking forward to speaking with colleagues contributing to that, so we can learn from each other.
FLG: And what would you hope attendees take away from the event itself, and specifically from your talk?
Anas: My talk focuses on the practical use of multi-omics—how it can be applied to advance new therapies and better understand disease systems biology.
It brings together previous work from the literature, along with guidelines and benchmarks, to show how multi-omics works in practice when deployed in biomedical research at Novartis.
FLG: Thank you so much again for making the time to speak to me today, and I'm really looking forward to hearing your talk at The Festival.
)
)
)