Blog

At Nucleai, we appreciate AI every day. But on July 16, we join the many other AI enthusiasts across the world in celebrating #AIAppreciationDay. One of these remarkable leaders is Suzana Couto, DVM, PhD, Dipl. ACVP, who serves as Senior Vice President of Pathology and Toxicology at Neomorph.

Suzana has over 15 years of experience leading teams and working in multiple stages of drug discovery, including target selection, pre-clinical research, nonclinical safety, translational development, and companion diagnostics. She remains at the forefront of digital and computational pathology and is passionate about leveraging AI in medical imaging.

From our conversation, it was clear that we need to appreciate the impact of AI not only on spatial biology, but also on every aspect of therapeutic development.

Q: What sparked your interest in AI and digital pathology?

Suzana: My journey into AI began during my Ph.D. in comparative oncology, when computational pathology was still in its infancy. I was fascinated by digital pathology and computer vision, which led me to beta test digital scanners and analyze quantitative data from images.

I remember when the first whole-slide scanner came out. Prior to that, I was just looking at photographs and measuring things very manually. But with whole-slide scanning and later AI-powered analysis — I just fell in love with the possibility of getting more quantitative data and spatial relationship data out of the images. And that’s how I started with AI early in my career as a pathologist.

Q: How has AI influenced your career path?

Suzana: I think a pivotal moment was when, about 6 years ago, I got involved in a large AI initiative, aiming to digitize many aspects of translational research and development. As part of this initiative, I had the chance to collaborate with a multidisciplinary team across multiple companies and academic collaborators. One of those companies was Nucleai, with whom we did a proof-of-concept digital pathology project. Nucleai had a particularly flexible and creative approach, was able to adapt their tools to our images and processes, and work in a very collaborative and interactive way with us to solve problems.

After that, we started working together in several projects, mostly in the translational space for predicting biomarkers for our clinical trials. I was also involved in initiatives to evaluate AI companies working with patient surfacing. Integration of big data, and natural language models, so, AI took me beyond digital pathology to far broader applications.

Q: How is AI revolutionizing the drug discovery & development process?

Suzana: AI has accelerated molecule design, protein and chemical structure prediction, and patient response modeling, and so much more. I’ll give 3 examples:

1. Data interrogation: One example of an area where AI has had an outsized impact is in the early clinical phase, when you’re getting ready to get a program launched in the clinic. Every company has a lot of data, including preclinical and translational data as well as data generated during Phase I trials, that can be a rich source of insights. But it’s really resource-intensive for companies to organize and mine their own internal databases. Companies end up with data scattered in different data rooms and dashboards, and in servers that are only available to a few individuals or departments.

With the emergence of cloud computing, the implementation of smart and secure data lakes, and the launch of generative AI, companies started training their own models to shorten the time between finding data and generating insights. AI can shorten the timelines in drug discovery by integrating large data sets and making them available to relevant stakeholders, fostering collaboration. Before, teams sometimes wouldn’t even be able to find useful data because it would have been archived in different siloed servers.

2. Data integration: AI has also revolutionized data integration in early clinical research, when you’re receiving a lot of data from patients, such as safety data, clinical response data, and exploratory biomarker information. New AI models can gather all this fragmented data plus real-world evidence and data from past trials to model response and help with reverse translation. AI is helping the drug development teams to make decisions that inform adaptive trial design in near-real time, with visibility into how the patients are doing across multiple criteria.

3. Spatial biomarker development: Spatial biomarkers have been shown to be predictive of clinical response in some cases. This goes for example to PD-L1 expression distribution or infiltration of cytotoxic T cells in tumors. But scoring spatial biomarkers visually can be hard and prone to variability. What I’m seeing is a push to use digitized images, combined with AI-based analysis to generate the readout in a much more scalable, reproducible way. Powered by AI, spatial biomarkers have the potential to reshape clinical trials by empowering drug development teams to accelerate patient enrollment, optimize dosing, and ultimately get the right molecules to the right patients, faster. There are also ongoing efforts by multiple institutions towards mapping all molecules of the human body, in health and disease, with focus on bringing spatial technology applications to the broader medical research community. A great example of that is HuBMap effort and some of the NIH funded innovation labs, designed not just to develop new assay technologies for spatial biomarkers, but to make the necessary computational power available to the clinic.

Q: What challenges does the industry face in adopting AI?

Suzana: One major challenge is the siloed nature of AI-powered solutions currently being built by isolated academic organizations, individual biopharma companies, and technology service providers. Many individual groups are working really hard on applying AI to specific problems at hand — but those solutions often stay siloed as the property of their creator or client once the project is over.

There’s a need for pre-competitive collaboration to make solutions more broadly available, rather than just for one drug or one population. But for collaboration to succeed, we will have to work with resource limitations, operational challenges, and regulatory requirements. For example, innovators in academia are working hard to move AI-powered diagnostics into the clinic. They are constrained by budgets and computational power needed, and will need collaboration from commercial organizations. As another example, medical device development teams have fewer operational constraints, but are concerned that the FDA may not accept data generated using AI-enabled technologies if submitted as part of a trial.

Q: How can collaboration help overcome these challenges?

Suzana: Cross-functional collaboration is essential. We need larger organized task forces involving regulators, scientists, industry, and medical professionals to address regulatory hurdles, operational costs, and resource requirements. The goal is to benefit patients by implementing AI in clinics, which requires policy changes and insurance company involvement for reimbursement.

I’m encouraged by projects such as the Digital and Computational Pathology Tool Harmonization Project, where academic centers, hospitals, pharmaceutical companies, algorithm developers, and advocates are collaborating to improve the reproducibility of biomarker measurements across digital pathology platforms.

Q: What is your vision for the future of AI in drug discovery & development?

Suzana: I envision an end-to-end AI-assisted ecosystem that streamlines every step of drug development and medical care, enhances patient diagnosis, and accelerates treatment delivery. To start, we can use AI to mine electronic medical records and generate epidemiological data to understand the journey of patients, and identify gaps and opportunities for intervention. Building the necessary infrastructure and leadership for this ecosystem will be key to driving progress and reducing the time to develop and deliver drugs to patients.

Also, AI can greatly streamline computation to enable virtual experimentation. For example, I imagine identifying biomarkers in small biopsy samples, such valuable pieces of tissue, that won’t even need to be stained — after simple microscopy to detect tissue autofluorescence and light reflection, AI models could generate virtual stains of every kind for every kind of marker to produce a huge amount of information in very little time with minimal resources.

As you can see, I enjoy working not only on very focused AI applications, such as digital pathology, but also on projects influencing the broader impact of AI on healthcare.

Thank you Suzana — we’re thrilled that you’ve joined us on the journey through the AI revolution.

Suzana’s outlook on AI reminds us that, often, technology developed to solve focused problems ultimately can have a global, generalizable impact — and it’s likely that AI itself will be a tool for identifying additional use cases for the AI-powered tools we are building today at Nucleai. On this AI Appreciation Day, we are grateful for the chance to pause and consider the exciting potential impact of AI-driven spatial biomarkers, not only in drug development but across the entirety of healthcare.