How ConvergeCELL’s Virtual Cell identifies clinically relevant biomarkers and therapeutic targets—without indication-specific fine-tuning.

Modern biomarker discovery is often constrained by narrow indication-specific models, limited datasets, and long iteration cycles. In this case study, we demonstrate that ConvergeCELL can robustly identify established biomarkers and prioritize novel targets across multiple disease areas using a true zero-shot approach, even when data is sparse and heterogeneous.

The challenge

Biomarker and target discovery becomes increasingly difficult as programs move beyond well-characterized indications. In practice, teams face:

• Highly variable cohort sizes across disease areas

• Noisy or incomplete clinical datasets

• Limited opportunity for extensive model tuning per indication

• Pressure to identify both known benchmarks and novel candidates early

The key question we set out to answer was: Can a single, indication-agnostic system reliably recover known biology while surfacing new insights across diverse disease states, without disease-specific optimization?

Objective

Evaluate ConvergeCELL’s “Virtual Cell” capability to identify:

• Established clinical biomarkers

• Known therapeutic drug targets

• Putative novel biomarker candidates

across multiple, biologically distinct disease states using a zero-shot workflow, with no indication-specific fine-tuning.

Study design and methods

To reflect real-world conditions, we curated patient cohorts spanning four disease areas with deliberately varied dataset sizes:

• Crohn’s disease: 150 disease / 75 control

• Lung cancer: 171 disease / 21 control

• Diabetes: 5 disease / 6 control

• Parkinson’s disease: 6 disease / 6 control

Analytical workflow

1. Zero-shot application of ConvergeCELL

   ConvergeCELL was applied without any indication-specific training or tuning.

2. Prioritization via ConvergeCELL score

   Outputs were ranked using a ConvergeCELL-derived scoring framework.

3. Benchmarking and validation

   Results were compared against:

   • Standard Differential Expression (DE) approaches

   • Established biological knowledge from Open Targets

4. Automated literature validation

   ConvergeCELL’s automated literature review system was used to:

   • validate nominated novel candidates

   • generate potential mechanisms of action

   
   This approach mirrors how discovery teams operate under real clinical constraints.

Results

Across all four disease indications, ConvergeCELL delivered consistent and robust performance.

Key outcomes

• Successfully identified established biomarkers and known therapeutic targets across all indications

• Identified novel biomarker candidates supported by literature evidence

• Outperformed differential expression (DE) approaches in all four indications

Demonstrated improved ranking quality, including:

• NDCG

• Kappa

• Gamma scores
  
  These metrics confirm that ConvergeCELL’s prioritization aligns closely with existing scientific literature—while extending beyond it to propose new hypotheses.

Example: diabetes biomarker prioritization

In the diabetes cohort, ConvergeCELL ranked top diabetes-associated genes while simultaneously distinguishing:

• established clinical biomarkers, and

• novel candidates supported by hypothesis generation and literature signals

Notably, ConvergeCELL correctly identified insulin as the most significant biomarker despite the absence of indication-specific tuning, highlighting its ability to recover core disease biology even in small datasets.

Why this matters

This study demonstrates that ConvergeCELL is not optimized for a single disease or benchmark; it is built to generalize across biological systems.

ConvergeCELL enables discovery teams to:

• Confidently prioritize biomarkers and targets across multiple indications

• Extract value from limited or uneven clinical datasets

• Reduce dependence on manual curation and disease-specific modeling

• Accelerate early discovery while maintaining biological credibility

In short: ConvergeCELL performs where real-world discovery happens, not just in idealized datasets.

Conclusion

ConvergeCELL has demonstrated that it is a reliable, indication-agnostic engine for biomarker and target discovery, capable of operating effectively across a broad spectrum of disease states, even when data is limited and indications are novel.

This capability is essential for modern discovery programs seeking speed, scalability, and biological rigor.

Click here to learn more about ConvergeCELL™.
Click here to contact the Converge team.