AI-Driven Spatial Biomarkers: The Next Frontier in Precision Medicine

Home » AI-Driven Spatial Biomarkers: The Next Frontier in Precision Medicine


Unlocking the mysteries of cancer goes beyond conventional diagnostics. In studying the complexities of tumors and therapeutic response, oncologists face a pivotal challenge: how to decipher the intricate landscape of cellular interactions and disease pathways.

Traditional biomarkers, once heralded as the beacon for targeted therapy, fall short in providing a comprehensive understanding of the tumor microenvironment. This challenge sets the stage for the convergence of spatial biology and artificial intelligence (AI) in reshaping the future of oncology.

Challenges with traditional biomarkers

The pathologist’s early role resembled that of a detective, relying on visual cues from surgeons regarding tumor size, grade, and lymph node involvement. However, this approach had limitations. Visual cues alone often failed to capture the nuances of the disease.

With the emergence of biomarkers, a new era has dawned in the field of oncology. These molecular signposts allow for a more precise understanding of disease behavior and the potential to tailor treatments to patients. Biomarkers provide critical insights into resistance mechanisms, aiding clinicians in adapting their therapeutic strategies.

While targeted therapies depend on the identification of the target mutation using next-generation sequencing (NGS), immunotherapies depend on PD-L1 scoring, which, despite its widespread use, only demonstrates a response rate of 20-30%. This stark discrepancy underscores the limitations of PD-L1 as a biomarker, particularly its inability to adequately reflect the complexities of the immune system within the tumor microenvironment, a critical factor in determining treatment response.

This pivotal realization has marked the genesis of a new challenge in oncology:  How can we identify effective predictive biomarkers for immunotherapy that offer deeper insights into the tumor microenvironment (TME), thereby enhancing our understanding of its role in predicting the response to immunotherapy?

The critical role of spatial biology

Spatial biology has become a crucial tool for unraveling complexity in cancer research. Moving beyond traditional genetic and proteomic approaches, the focus has shifted towards understanding the spatial relationships within the TME in order to realize deeper insights into cancer biology.

Spatial biology delves into the spatial organization of the tissue microenvironment, offering a three-dimensional perspective on tissues and their components. Through the lens of spatial biology, we can develop an understanding of how cells interact within their microenvironment, shedding light on the interactions between tumor cells, immune cells, and stromal elements. The additional context is paramount in deciphering the behavior of these cells and their responses to therapeutic interventions.

A meta-analysis performed by Lu et al. sought to determine which biomarker modalities were most predictive of response to checkpoint inhibitor therapy in non-small cell lung cancer. Their analysis revealed that a combination of multiplex immunofluorescence (mIF) with other modalities was significantly more predictive of response compared to other methods.

Unlike traditional biomarkers that might provide a global assessment, spatial insights enable the identification of distinct neighborhoods within a tumor, each with its unique characteristics. This granularity is vital, especially in the era of personalized medicine, where a one-size-fits-all approach proves increasingly inadequate.

Uncovering the spatial nuances of the TME grants researchers and clinicians a more comprehensive understanding of how treatments may impact different regions within a tumor.. This insight is particularly crucial in immunotherapy, where the interaction between tumor and immune cells plays a pivotal role in treatment outcomes.

Combining spatial biology with artificial intelligence only amplifies its impact, allowing researchers to analyze vast datasets and extract meaningful patterns. This integration could potentially transform our understanding of cancer, moving us beyond traditional biomarkers towards a more comprehensive grasp of the cellular interactions and relationships influencing disease progression and therapeutic response.

Case study: Decoding immunotherapy response with AI-driven spatial biology

Currently, non-small cell lung cancer patients with PD-L1 expression greater than 50% receive pembrolizumab as primary therapy with a fairly good response. However, within this PD-L1 positive population, a subset of patients remains unresponsive.

Results published at ASCO 2021 indicated that while PD-L1 scores were similar in both responder and non-responder groups, spatial analysis of tumor-infiltrating lymphocytes (TILs) in the TME could be used as a predictive biomarker for patients who would benefit from pembrolizumab.

A series of tissue images were taken from high-expressing PD-L1 patients (Figure 1, top). Both the responder and non-responder samples suggest TILs associated with tumor cells (in blue), offering no immediate visual distinction.

Figure1: Feature calculation of TIL distance from tumor core in NSCLC patients, with orange being high proximity and purple being low proximity. Figure 2: KM analysis- positive patients had a significantly higher median OS (NR vs.17.8m, p<0.05) and 2-year OS
(70.8% vs. 33%, p=0.02) than negative patients.

After using deep learning models to calculate the map the proximity scores of each TIL to tumor cells, heat maps (Figure 1, bottom) were created, showing that responders had TILs in very high (orange) proximity to the tumor, while non-responders had a mix of TILs in high and low (purple) proximity to the tumor. This subtle yet impactful difference is key to distinguishing responders from non-responders within this high PD-L1 cohort.

AstraZeneca is using a similar type of analysis to study response to antibody drug conjugate therapy for Her2-positive breast cancer. They’ve developed a “spatial proximity score” which measures the percentage of cells which are Her2-positive or have at least one HER2-positive neighbor within 50 µm–a measurement impossible for pathologists to discern with the naked eye. High spatial proximity scores have been predictive of improved treatment outcomes.

The integration of AI-driven spatial analysis adds a new layer of information to tissue samples. By discerning patterns within the spatial context of the tumor microenvironment, AI promises a more nuanced and personalized understanding of treatment responses, paving the way for enhanced clinical outcomes.

AI spatial analysis redefines clinical trials and diagnostics

AI is spearheading a revolution in spatial biology, reshaping the landscape of research with the potential to deliver improved patient outcomes. AI-driven spatial biology has a wide variety of applications, including:

  • Translational Research: AI-driven  spatial biomarkers can significantly advance translational research by aiding pharmaceutical companies in generating and validating hypotheses, thereby enhancing the understanding of the Mechanism of Action (MoA) of their drugs.
  • Clinical Development: AI-driven  spatial analysis can streamline patient eligibility assessments for prospective clinical trials. AI can lead to efficiency gains achieved by predicting molecular markers at the time of diagnosis, expediting the enrollment process.
  • Companion Diagnostic Development: Spatial predictive biomarkers can offer actionable insights that may guide diagnostic and therapeutic decisions.
  • Diagnostic Augmentation Tools: AI-driven  spatial analysis can improve the reproducibility and accuracy of biomarker scoring between pathologists.

Beyond these applications lies a broader vision–the era of “health intelligence”. While precision medicine focuses on extracting features from biological data, health intelligence takes a giant leap forward. It involves the integration of genetics, phenotypes, x-rays, histology slides, and clinical records, utilizing machine learning and computation to derive profound insights. Nucleai provides cutting-edge solutions to help shape the future of healthcare.

Book a demo to explore Nucleai’s spatial biology solutions in action.

Lu, S., Stein, J. E., Rimm, D. L., Wang, D. W., Bell, J. M., Johnson, D. B., … & Taube, J. M. (2019). Comparison of biomarker modalities for predicting response to PD-1/PD-L1 checkpoint blockade: a systematic review and meta-analysis. JAMA oncology, 5(8), 1195-1204.

Book A Demo

Add Your Heading Text Here