Composite intelligence: How AI-driven biomarkers are setting the bar in immuno-oncology

Original source here

‍

A new chapter for immunotherapy decision-making

Immune-checkpoint inhibitors (ICIs) transformed treatment paradigms in dozens of malignancies, yet oncologists still confront a deceptively simple question at every consult: Will this individual patient benefit and at what cost?

Conventional, tumor centric assays such as PD-L1 immunohistochemistry, tumor-mutational burden (TMB), and next-generation sequencing (NGS) provide valuable clues, but they were never designed to explain the full, fast-moving conversation between tumor and host. They cannot, for example, capture the rapid cytokine surges, innate-cell recruitment and metabolic rewiring that unfold within hours of the first infusion and ultimately shape clinical outcome.

“Therapies have sprinted ahead, while our biomarkers still jog,” observes Ofer Sharon, MD, CEO at OncoHost. “We now need biomarker strategies that look at the patient’s systemic reaction – because that reaction is where immunotherapy succeeds or fails. If we ignore the patient’s systemic reaction, we are mapping only half of the battlefield”.

Host response: The missing dimension

Blood, unlike tissue, is a living transcript of the whole-body dialogue. Every vial contains not only circulating tumor DNA but also thousands of proteins secreted by immune and stromal cells – real-time indicators of activation, suppression, stress and repair. High-throughput proteomic platforms now quantify these signals at industrial scale, turning a routine blood draw into a systems-level snapshot of immune engagement.

“Proteins are the immune system’s real-time dial tones,” explains Michal Harel, PhD, VP of Translational Medicine at OncoHost. “Measuring enough of them at once reveals hidden interactions and patterns that DNA or single-protein assays simply can’t capture.”

Feed that data into modern machine-learning pipelines, and a composite “signature” emerges – compressing multidimensional biology into an actionable score.

From raw signals to actionable scores: The rise of AI-composite biomarkers

Consider a patient with advanced NSCLC whose tumor lacks an actionable driver mutation. The therapeutic pivot is immunotherapy, but which regimen – monotherapy, chemo-immunotherapy, or a trial combination – offers the best odds?

Proteins are the immune system’s real-time dial tones. Measuring enough of them at once reveals hidden interactions and patterns that DNA or single-protein assays simply can’t capture.

The sheer dimensionality of proteomic data demands modern machine learning. By recognizing non-linear interactions among hundreds of markers, AI transforms raw volatility into a single probability score that clinicians can interpret in seconds. Composite host-response signatures therefore add an entirely new decision layer without discarding the old—PD-L1 and NGS remain useful for eligibility; AI-powered composites refine how to use that eligibility in practice.

A case in point is a plasma-based signature that analyses a pretreatment blood sample, integrates the patient’s PD-L1 value and outputs the likelihood of durable benefit from ICI monotherapy versus chemo-immunotherapy. In a prospective, multi-centre study published in JCO Precision Oncology, the score re-stratified roughly one-third of advanced NSCLC cases into a different first-line regimen and translated into a measurable survival advantage. A blinded decision-impact survey later showed clinical traction – 93% of oncologists adjusted at least one recommendation after seeing the data, underscoring how quickly composite intelligence translates to bedside practice.

“When clinicians see a broad host-response signal distilled into a traffic-light output, it becomes a decision layer they don’t want to practice without,” notes Yehonatan Elon, PhD, CTO at OncoHost. “It moves the conversation from population-level odds to patient-level insight.”

These signatures emphasize complementarity, not competition: tumor assays spotlight the disease, while proteomic composites spotlight the patient’s real-time capacity to fight it.

Extending precision to safety: the irAE frontier

Efficacy is only half the precision equation. Severe immune-related adverse events (irAEs) can derail, force discontinuation or cause long-term morbidity. Composite host-response models such as PROphetirAE™, are now being trained to predict irAE risk before the first infusion. This composite biomarker analyses pre-treatment plasma to flag patients at high likelihood of developing grade 3/4 immune-related adverse events before therapy even start. As validation matures, such a tool could prompt intensified monitoring, steroid prophylaxis or an adjusted dosing schedule before toxicity strikes.

Sharon summarises the two-pronged strategy: “One composite score guides how hard we push; another warns how hard the immune system might push back. That is genuine precision care.”

Regulatory and market momentum

AI-driven composite biomarkers are maturing at exactly the moment regulators and payers are demanding more robust, real-world evidence. Regulators now expect every algorithm to ship with a “maintenance manual” – locked code, auditable version history, and built-in drift surveillance, while insurers prioritize tools that reduce futile therapy and curb high-grade toxicities.

When an oncologist can open the chart and view a living, data-rich profile – efficacy probability, toxicity risk, resistance trajectory, ‘precision medicine’ stops being a buzzword,” Sharon concludes. “It becomes the operational standard of cancer care.

Against that backdrop, a next-generation assay must:

• Fuse tumor-centric and host-immune signals into a unified read-out.
• Leverage AI to compress thousands of variables into a single, clinician-friendly index.
• Refresh dynamically as treatment reshapes biology, supporting adaptive management.
• Complement rather than compete with established markers such as PD-L1.

Early composite platforms – for example, those predicting treatment efficacy or pre-empting immune-related adverse events, demonstrate just how powerful the marriage of deep proteomics and disciplined machine learning can be. They are not niche curiosities; they foreshadow the new normal for biomarker development and signal where the field is headed.

A future built on composite intelligence

Biomarker science is pivoting from single-analyte snapshots to systems-level, AI-powered composites that evolve with the patient. The next five years will see these signatures embedded in adaptive trials, electronic health-record dashboards and payer algorithms. Academic investigators will deploy them to enrich study cohorts; community oncologists will use them to tailor first-line regimens; regulators will reference them when defining “clinically meaningful” benefit; and payers will lean on them to justify value-based contracts. Doing so will close the gap between therapy potential and real-world outcomes.

“When an oncologist can open the chart and view a living, data-rich profile – efficacy probability, toxicity risk, resistance trajectory, ‘precision medicine’ stops being a buzzword,” Sharon concludes. “It becomes the operational standard of cancer care.”

AI-powered composite biomarkers – once a bold concept – are rapidly solidifying into the new gold standard. Those who master host-response science today will define what “next-generation” means tomorrow.

‍