In my years working across life sciences, one question comes up again and again: What’s next for AI in our field? The truth is that the life sciences industry faces challenges unlike any other.
Where a bank or retailer might deploy AI chatbots to improve customer service, our world is defined by enormous, messy datasets, including clinical trials, lab results, publications and patient records. This must be interpreted with care. The stakes are not just efficiency or convenience; they are breakthroughs in treatment, safety and patient outcomes.
That’s why I believe the real opportunity for generative AI (GenAI) in the life sciences is not in chatbots, but in enabling deep and precise retrieval. Success here means connecting across multiple sources, reconciling heterogeneous data and surfacing insights that a human researcher would struggle to piece together.
Imagine asking: “Find me colorectal cancer trials using ZALTRAP [a drug] with the most recent supporting publications.” GenAI, when applied effectively, can handle that complexity, and this is where the next frontier begins.
From Traditional Search to AI-Driven Discovery
For decades, search in life sciences has mostly meant keyword lookups or rule-based retrieval. Researchers, clinicians and pharma teams relied on these tools to sift through scientific literature, clinical trial data, patents and regulatory filings. They worked well enough for simple, well-defined questions. But as soon as you needed to account for domain-specific language, synonyms or the complex relationships between diseases, molecules and pathways, traditional search hit its limits.
The result? Endless manual refinements, stitching insights together from different sources and lots of time spent just finding the right information. Reac
You can ask complex, natural-language questions and get results that connect the dots across literature, trials and patents — even when they use different terminology.
Now, with GenAI and large language models (LLMs), that’s changing. LLM-powered search understands meaning, not just exact words. You can ask complex, natural-language questions and get results that connect the dots across literature, trials and patents — even when they use different terminology. This opens up entirely new ways of working: identifying drug repurposing opportunities hidden in disconnected studies, accelerating biomarker discovery or finding previously unseen links between biological entities. It’s faster, more comprehensive and far less manual.
Why Tensors Matter in This Shift
Life sciences data comes in all shapes and sizes — omics data, 3D protein structures, medical images, regulatory documents, clinical trial reports and more. Most of it is unstructured or semi-structured, which makes it tricky for AI systems to find and assemble relevant information quickly. Given the nature of life sciences, accuracy is critical. “Good enough” seldom exists.
This is where tensors come in.
So, what is a tensor? Think of it as a multidimensional data container. A vector is a one-dimensional list of numbers. A matrix is two-dimensional. A tensor goes beyond that, capturing multiple dimensions at once. This allows AI models to represent complex relationships — like spatial configurations of proteins or contextual relationships between words in a scientific article — even if those pieces of information are far apart.
In other words, tensors let AI “see” and learn patterns that are deeply embedded across different dimensions of data.
Tensors in Action: Protein Structures
Take structural biology as an example. Models like AlphaFold use 3D tensors to represent the spatial relationships between amino acids. These tensors allow the model to learn how proteins fold, twist and interact — crucial knowledge for understanding disease mechanisms and designing new therapies.
When you embed a protein as a tensor, you preserve:
- Sequential data (the order of amino acids)
- Spatial relationships (how parts of the protein fold and connect)
- Biochemical properties (like charge or hydrophobicity)
This rich representation lets machine learning (ML) models predict protein folding, identify binding sites, map protein-protein interactions and even discover new drug targets.
The same idea applies beyond proteins.
Medical imaging, for example, can use tensors to encode not just pixels, but also their contextual relevance, helping AI detect subtle cancer markers even in noisy scans. In clinical settings, tensors help AI analyze data streams from wearables or Internet of Things (IoT) devices in real time, enabling faster interventions.
Beyond Retrieval: AI Agents in Life Sciences
AI agents are another emerging application. Think of them as intelligent assistants that continuously gather, analyze and synthesize information across fragmented data sources. An AI agent could monitor new literature, clinical trials and regulatory updates in real time, summarize findings and even suggest next research steps.
Good agents don’t just fetch information — they connect it, building context and reason through problems step by step.
The key here is multistep reasoning. Good agents don’t just fetch information — they connect it, building context and reason through problems step by step.
This means faster reasoning, better accuracy and more meaningful insights. This allows you to stitch together multimodal data and ask questions across modalities and time. For example, as illustrated in the example below, you can now find patients for trial recruitment for a disease subtype based on certain progression (or regression) images over time. You can do this by combining into a tensor the patient’s medical record, biomarker assays, histopathology slides and any other prognosis outcome notes.
Why This Matters
Life sciences are moving into an era where data is simply too complex and too vast for traditional tools. Tensors provide the foundation for AI models to handle this complexity, enabling everything from better search to advanced reasoning. Whether it’s predicting protein structures, extracting insights from clinical data or powering AI agents that help researchers focus on discovery rather than data wrangling, tensors are quietly becoming the backbone of the next wave of AI in life sciences.
YOUTUBE.COM/THENEWSTACK
Tech moves fast, don’t miss an episode. Subscribe to our YouTube
channel to stream all our podcasts, interviews, demos, and more.
