Intelligent Omics' Dr Rob Grundy discusses how combining AI with big data can lead to better drug discovery processes and that partnerships can help more organisations leverage cutting-edge tools.
Dr Robert Grundy MBE has over 25 years of experience in pharma and biotech, including leadership roles at Schering-Plough, GSK, Cerebricon, and Almac Discovery. He is CEO of Intelligent OMICS, an AI-driven drug discovery company, and has advised government policy on science and technology. His keynote, ‘Exploiting big data to drive drug discovery through AI and collaboration’, will highlight the role of AI in accelerating innovation.
Dr Robert Grundy MBE has spent more than 25 years in Pharma and Biotech beginning in drug discovery and development at Schering-Plough and GlaxoSmithKline. Dr Grundy served as CSO with Cerebricon before joining Almac Discovery in 2008.
2014, Dr Grundy founded Anglezarke Life Sciences, a commercialisation and growth strategy consultancy. He has served as CEO of CIGA Healthcare and Health Innovation Research Alliance Northern Ireland. He has held or holds visiting and honorary lectureships at King’s College London and Queen’s University Belfast.
Dr Grundy formally chaired Matrix, the Science and Technology policy panel for the Northern Ireland Government and represented Northern Ireland on the chief scientific advisors Network for the UK. He previously served as CEO at Intelligent OMICS, an AI-driven drug discovery company, where he retains a position as a senior advisor. Dr Grundy now works as Director of Innovation Partnerships at Catalyst. Dr Grundy’s keynote presentation at ELRIG Research & Innovation 2025 will focus on ‘Exploiting big data to drive drug discovery through AI and collaboration’.
Can you tell me about yourself and your role at Intelligent Omics?
Grundy: My role within Intelligent Omics has been to use a powerful machine learning-driven approach to analyse large data sets gathered from disease patients. This computational approach allows us to understand biology better. If we understand biology and how disease biology differs from healthy biology, then that will help us identify better targets for drug discovery. The alternative is iterative peer-reviewed research into disease biology, which we've always done. But it is fundamentally flawed because it's open to bias and the non-publication of negative results.
The proof that it's inefficient is the significant failure rate of experimental medicines in the clinic, which is to the detriment of patients, pharmaceutical companies, and investors in those pharmaceutical companies. The whole thing is grossly inefficient and needs to be improved. If we can use data and the cost of computation in months or weeks rather than years, we can present an alternative view of biology. This can reveal dark biology that was previously unappreciated and provides a meaningful alternative approach to developing new drugs.
How does AI integrate into this field of research?
Grundy: Analysing the biology and the processes involved in drug discovery has produced a huge amount of data. The execution of healthcare also produces a massive amount of data. The question is, how do we employ that data to develop better drugs?
An increasing number of tools will allow us to do that. As AI becomes more sophisticated and pursued by a larger number of organisations, there will be an overwhelming number of tools. Undoubtedly, this will allow us to better use the data that we generate. My hypothesis for this presentation is that the tools are out there, but how can large and small organisations access them efficiently?
I suggest that partnerships allow us to do that in a cost-efficient way that doesn't expose us to investment in technologies that we may only need from time to time. The learnings achieved by using these tools against the big data we're producing across disease biology can be shared and learned within collaborative environments. In that way, you increase the menu of tools available to you.
My observation as a service provider in this area is that once an organisation invests in internal AI or machine learning capability, it prevents them from engaging with additional solutions outside of the area in which that investment has been made now. In a time when technology is moving so fast, that doesn't make sense. You should always be able to take advantage of new technology as it emerges, for the purposes that you understand it will add value.
If you're prevented from doing that because of the investments made, you're unable to invest in leading-edge technology, and that won't benefit the organisation. There has to be a more efficient way to access evolving technology that reduces risk exposure and induces reduced expenditure.
How can organisations best employ AI for biological data?
Grundy: We must understand where the value can be utilised and where more traditional methods might be better employed. It will always be a hybrid approach. AI will be transformational. We just don't know how yet. It's bound to be an iterative process of using AI and comparing it with traditional methods to ensure you've augmented what you do traditionally rather than replaced it.
It's all about augmentation. You can't just throw away years and years of iterative experience in any process by replacing it with AI; that's just not going to work. I don't believe anybody's doing that, but you must take a sensible and diligent approach.
If you want to do something new by employing an AI tool, make sure that you expose it to a level of diligence by comparing it to your traditional methods, understanding how it's improved, and understanding where the complementarity is. But that requires effort, and so it's counterintuitive. If you're going to do it properly, you probably won't reduce the effort level, but you will increase the quality of what you do.
What are the challenges? Is data quality a challenge for AI?
Grundy: Data quality is certainly a consideration, a factor. I'm not going to say it's a challenge or an issue because I don't think it is. It's a part of the process. The process begins with identifying data sets. The second stage is quality control of those data sets. The third stage is either abandoning the data set because its quality isn't there or subjecting it to treatments that allow it to be used.
We use principal components analysis and TSNY plots, which allow us to understand how data is distributed. Then, we make a decision on whether this is a reliable data set. Data quality will always be a consideration, and that's where you start, right? If you know your technique well, you will know what you require of the data and how to treat it to make it appropriate for use.
Duplication of effort is a significant issue in drug discovery. Part of that is the necessity to be competitive. You don't want to share that you've got data with another organisation trying to achieve the same aim. But there's a certain tragedy if you're trying to cure a disease with an unmet medical need. That is something we wrestle with in the drug discovery and development industry as a whole, which is the duplication of effort that can be avoided to make medicine development more efficient.
You mentioned bias, how does that impact AI research using biological data?
Grundy: Initiatives such as AllTrials, initially driven by Ben Goldacre, compel clinical research practitioners to publish all their negative or positive results. So, that's getting better. Bias is inherent in published medical research, but it's also inherent in the AI tools that are developed. For example, take AlphaFold from DeepMind. Everybody celebrates that as a great way to understand complex protein structures and how we might target them, but it's built on a biased data set. It's built on the available data, which is largely positive, and that means that there are gaps in that data that prevent you from seeing the whole picture. A tool like AlphaFold will never be perfect until you get a significant quorum of of negative data to complement the positive data from the groups that are adding to the data sets from which things like AlphaFold are built.
If you go back to bias in drug development, the best example, to my mind, is Alzheimer's disease. So, in Alzheimer's disease, for probably two or three generations of medical research, we've pursued two very narrow hypotheses, the amyloid and the tau hypotheses, which have consumed billions of research dollars. We still don't really have a meaningful cure for Alzheimer's disease or understand how it works.
Increasingly, data is telling us that this is a disease of cellular metabolism of inflammation and characteristics like amyloid and tau are probably only manifestations of the disease, not causes of the disease. But because so much has been invested in those narrow hypotheses, it's prevented more investigation of the of the other elements of disease.
I think that's a real problem that is a product of peer review. The problem with peer review as a process to discover drugs is that research gets funded based on what's published. Data gets published based on what everyone agrees is good quality data. If the majority of the people asked to agree on what makes good quality data have a certain interest in an area of biology, then everything moves in that direction. This can result in a mischaracterisation of the real situation in disease.
Reproducibility is another problem. Many projects seek to develop novel medicines based on experiments that haven't been appropriately reproduced. It's like setting the course for a rocket towards the moon. If you're off by a single degree, you'll miss.
It's the same developing drugs. Suppose you have biology that isn't represented in sufficient numbers in a clinical population. In that case, you'll never be able to get the effects you need in a clinical trial. Getting it right early and having the information you've analysed properly is really important.
Dr Rob Grundy MBE is the Director of Innovation Partnerships at Catalyst.
