Jonathan D. Grinstein, PhD, North American Editor of Inside Precision Medicine, hosts a new series called Behind the Breakthroughs that features the people shaping the future of medicine. With each episode, Jonathan gives listeners access to his guests’ motivational tales and visions for this emerging, game-changing field.
Precision medicine is often framed as imminent: gather more data, refine analytics, and individualized care will naturally follow. In reality, progress has been uneven. Genomic, imaging, pathology, and clinical data remain fragmented across systems and poorly integrated into clinical workflows. The core challenge is not data scarcity but the ability to interpret complex, heterogeneous inputs quickly enough to guide real medical decisions. To address this, Jurgi Camblong founded SOPHiA Genetics with a focus on building infrastructure rather than isolated tools—aiming to turn multimodal health data into actionable insights, a goal far more difficult in practice than in theory.
In Behind the Breakthroughs, Camblong highlights persistent structural and technical barriers limiting data-driven healthcare. Genomic standardization, for example, remains inconsistent, with approaches ranging from targeted panels to whole-genome sequencing, each balancing cost, sensitivity, and speed. The field is also shifting from single mutations to complex interactions among variants. Expanding beyond genomics adds further complexity, as transcriptomics, radiology, liquid biopsy, and computational pathology each involve distinct methods and clinical uses. Rather than enforcing uniformity, SOPHiA Genetics works across this diversity to produce consistent, clinically usable outputs despite technological and regulatory variation.
Ultimately, success depends on integrating statistical, machine learning, and deep learning methods while staying grounded in biology. A major limitation is the lack of robust feedback loops: precision medicine requires long-term patient outcomes, which many systems fail to capture. Without this, even advanced models are constrained. The central challenge is execution—translating existing data into meaningful insights that improve individual patient care.
This interview has been edited for length and clarity.
IPM: What types of multi-omics datasets are currently workable and applicable in a clinical setting, and how do you see their role evolving in routine patient care?
Camblong: When we started in 2015 and launched the platform into the market, people were just analyzing CFTR for cystic fibrosis and BRCA1 and BRCA2, two genes for hereditary cancer. To be honest, there were some efforts around whole genome analysis, but it was very, very rare. Our intent was always not to be a research tool but a tool that brings real benefit to most patients routinely and safely, and things evolved over time.
Now, probably the mean number of genes analyzed when producing genomic information for a patient is around 100 genes. Then you have some solutions that require analyzing only 30 genes because you want to be extremely precise, cost-effective, and rapid. There are other solutions that require sequencing the whole genome. But getting full information with the same sensitivity you can have with smaller panels is not an easy task, and this is where algorithms are really important.
In our case, the fact that we have grown along this journey with the field gives us an advantage today, enabling people to produce more genomic information with the same sensitivity as smaller panels. Genomics is continuously evolving. In the past, people did not necessarily look at copy number variations. Now we are even talking about partial copy variations, like in a gene called PTEN, which is a driver gene, and where a partial CNV can be very important.
What I am trying to explain is that it is not yet simple. It is not streamlined. Lab protocols are different; sequencing approaches are different; it is a constant evolution. In our case, being an operating system that supports thousands of hospitals, we are privileged to be exposed to this complexity, which enables us to improve our algorithms more rapidly and deliver them back to users who can benefit from new capabilities.
Transcriptomics is becoming a very interesting data modality. Initially, it was used to detect so-called gene fusions, specific genomic features that are hard to detect from DNA and require RNA. I am quite bullish on transcriptomics. I believe it will enable cancer subtyping at scale, possibly with more efficient methodologies than what is done today on tissue. It may not replace tissue, but it may allow us to go further and, in some cases, provide more objective outcomes than staining protocols.
Along those lines, radiomics is also very important. By radiomics, I mean data produced by radiologists, CT scans, PET scans, and MRI. There is a signal in this data. For example, you can see if cells are necrotic. You get additional information based on tissue composition and imaging. You can automatically measure tumor volume.
In metastatic cases, where tumors are spread, measuring them is not necessarily easy. You can identify where tumors are, and this information, feature extraction from images, is very powerful. It is also the only data modality that is used longitudinally today in cancer to monitor response to treatment.
Another modality that will become important is liquid biopsy testing to follow patients longitudinally, based on molecular profiles and minimal residual disease (MRD). If you think about computational pathology, H&E staining in particular will be important. I am more skeptical about immunohistochemistry at scale, given feedback from pathologists; multiplexing may introduce too much signal and create confusion. Proteomics has potential, but clinically, it is not quite there yet. Even the most advanced actors are not fully at clinical utility.
Over time, we will need to combine these modalities and apply smart algorithms to extract signals and support decision-making. In the end, this is what matters: not computing data unless it brings value to the oncologist, pathologist, biologist, or geneticist.
IPM: How is the SOPHiA interface designed for clinicians in practice? What does the user experience look like across different use cases, such as oncology or liquid biopsy workflows?
Camblong: It is a web-based interface you log into. For example, if you are at Moffitt Cancer Center in Florida, using the platform for hematological malignancies, you will see which mutations are detected with high sensitivity and how actionable they are. If you are in a hospital in the U.K. using it for liquid biopsy testing, you will see the mutations identified for those patients.
We also have customers using it from a multimodal perspective, more from an oncologist’s point of view, where they can see how similar patients with similar molecular profiles respond to treatments elsewhere. For us, this includes partnerships with major clinical genomic databases. Through these, we provide access to additional data layers for institutions, even when the patient data originates locally.
The interface is always web-based. In the backend, we use microservices to compute data using AI, deep learning, machine learning, statistical inference, and pattern recognition. The user then leverages this information to make decisions and answer clinical questions.
IPM: Given the diversity of data sources and technologies, how do you approach standardization and harmonization across datasets, particularly in a global context?
Camblong: We operate in over 70 countries. We support local data production and management, but within a framework of collective knowledge. It is important to align solutions with regulations. In some countries, we operate in research mode only. In Europe, some applications are IVD, and in the future possibly In Vitro Diagnostic Regulation (IVDR) or companion diagnostic solutions.
The key is to build technology with optionality, documenting how it is built and its intended use. If you want to make clinical claims, you must conduct clinical studies. The foundation is design control, like in aviation, so that you ensure sensitivity, specificity, reproducibility, repeatability, and robustness, regardless of regulatory frameworks.
IPM: How does your platform adapt to the wide variety of user systems, including different sequencing instruments, workflows, and laboratory environments?
Camblong: The backend is fully engineered and automated. But workflows differ across hospitals due to global constraints and complexities. Managing this heterogeneity while delivering consistent outputs means adapting to different workflows. This is not easy, but we have demonstrated strong performance. For example, with Memorial Sloan Kettering, we accessed both their data and their applications, MSK-IMPACT and MSK-ACCESS. We industrialized these within SOPHiA without infringing on IP, enabling hospitals to produce data locally and leverage our algorithms. We achieved over 98% concordance across sites, comparable to repeating sequencing within a single workflow.
We also work with multiple sequencing vendors to ensure compatibility across instruments and consumables. Because we process large volumes of data, we can also advise on optimal workflows for specific applications. Since we are paid per use, our incentives are aligned with hospitals; better workflows mean more patient cases and better outcomes.
On AI: it is a toolbox. Different models suit different problems. Large language models are useful for text and sometimes images, but not everything. Understanding biology and data diversity is key to selecting the right mathematical model that scales effectively.
IPM: As you expand into adjacent domains like radiology, how do you approach entering new clinical areas while ensuring relevance and usability?
Camblong: Always with partners, healthcare institutions. We are strong in software, AI, and biology, but not medical practice. We co-develop with clinicians to ensure integration into workflows and real clinical benefit. For example, with MD Anderson, we collaborate on translational and routine lab work to move technologies into clinical practice, such as transcriptomics for cancer subtyping and MRD.
In multimodality, we work case by case. For instance, in kidney cancer in France, we partnered with the UroCCR network, analyzing 27,000 patient cases. This allowed us to identify signals and predict responses to immunotherapy. Innovation only matters if it is adopted in practice.
IPM: How actionable are your clinical decision-support tools today, and how do you incorporate real-time or longitudinal data?
Camblong: It depends on regulations. In some places, like the U.K., the platform provides information to oncologists, who then interpret it. For multimodality, feedback loops are essential, linking molecular data, treatment, and outcomes.
With UroCCR, we continuously improve algorithms using real-world data. We should be leveraging post-market data more systematically to refine treatment decisions. Real-world complexity can reveal which patients truly benefit from therapies. Longitudinal data is critical, not just for outcomes, but also for avoiding adverse effects. For example, some ovarian cancer patients benefit from PARP inhibitors but may develop leukemia. Understanding these patterns requires real-world data loops.
IPM: How do you think about data ownership, access, and control?
Camblong: Ownership does not exist in a strict sense. Individuals are the ultimate controllers. Hospitals and companies are processors. Data is critical for AI, but our model is decentralized: hospitals retain control of their data. Algorithms learn from data, but once trained, they can deliver insights without retaining raw data, enhancing privacy.
Also, oncology data does not age well because treatments and technologies evolve rapidly. What matters is continuous exposure to new data. Collective intelligence through networks and platforms is essential for precision medicine.
IPM: How does SOPHiA approach cross-border collaboration and democratization?
Camblong: Democratization means making technology accessible and usable. For example, in India, a hospital previously sent samples to the U.S., with high costs and six-week turnaround times. We enabled local testing within months, reducing turnaround to under two weeks and building internal expertise. This increased testing volumes and improved clinical adoption.
IPM: Are there areas less amenable to your approach?
Camblong: About 80% of our work is in cancer, 20% in rare disorders. Rare diseases require even more collaboration due to limited data. We support peer networks where clinicians share insights, for example, variant classifications, helping others make faster decisions. As medicine becomes more precise, collaboration becomes even more critical.
The post Jurgi Camblong: Data-Driven Doctors Without Borders appeared first on Inside Precision Medicine.
