COMPASS: A Web-Based COMPosite Activity Scoring System to Navigate Health and Disease Through Deterministic Digital Biomarkers

COMPASS is a deterministic, ontology-free web-based framework that converts gene expression into stable, interpretable pathway activity scores without requiring permutation or reference cohorts, offering improved robustness and discrimination compared to existing single-sample enrichment methods for precision medicine applications.

The Gist

Imagine you are trying to understand the mood of a crowded party. You have a list of 50 people (genes) who are either shouting excitedly (upregulated) or sitting quietly in the corner (downregulated).

The Old Way (Traditional Methods):
Most scientists today use methods like GSEA or GSVA. Think of these methods as taking a snapshot of the party and asking, "Who is louder than the average person right now?"

• The Problem: It's all relative. If the whole party gets louder, everyone looks "excited," even if nothing actually changed. If you take a photo of a different party later, the results might look totally different because the "average" changed. It's like judging a runner's speed by comparing them to other runners in the race, rather than checking their actual speedometer. It's also a bit of a guessing game that relies on random shuffling (permutations) to see if the results are real.

The New Way (COMPASS):
The authors of this paper, Saptarshi Sinha and Pradipta Ghosh, built a new tool called COMPASS. Think of COMPASS as a smart, automated traffic light system for your genes.

Here is how it works, broken down into simple steps:

1. Finding the "Red Light" (Thresholding)

Instead of asking "Who is louder than everyone else?", COMPASS asks: "At what exact volume does a gene flip from 'off' to 'on'?"

• Imagine a light switch. It's either off (0) or on (1). COMPASS looks at the data and finds the exact moment a gene's expression crosses that line. It doesn't care about the other genes; it cares about the specific "tipping point" for that specific gene. This is like setting a specific speed limit (e.g., 65 mph) rather than saying "drive faster than the guy next to you."

2. Measuring the Distance (Standardization)

Once the "Red Light" is set, COMPASS measures how far a gene is from that line.

• Is the gene just barely touching the light? Or is it screaming at the top of its lungs? COMPASS turns this into a simple score: "This gene is 3 steps past the limit." It does this for every gene, turning a messy, complex dataset into a clean, standardized score.

3. The Final Score (Aggregation)

Now, imagine you have a team of 50 genes. Some are "Good Guys" (they should be loud when the disease is active), and some are "Bad Guys" (they should be quiet).

• The Magic: COMPASS adds up the scores. If the "Good Guys" are loud and the "Bad Guys" are quiet, you get a high positive score (Disease Active!). If the "Bad Guys" are loud, you get a negative score.
• Why this matters: Old methods often get confused if a gene signature has both loud and quiet genes. COMPASS handles this perfectly, like a referee who knows exactly who is on which team and keeps the score accurate.

Why is this a Big Deal?

1. It's a "Digital Biomarker" (A Universal Translator)
Because COMPASS uses fixed rules (math) instead of random guesses, it works the same way everywhere.

• The Analogy: Imagine you have a recipe for a cake. If you use the old methods, the cake tastes different depending on who bakes it or what oven they use. With COMPASS, if you use the same ingredients (data), you get the exact same cake every time, whether you bake it in a home kitchen (a mouse model), a professional lab (a human organoid), or a massive factory (a human clinical trial).
• The Result: Scientists can finally compare a mouse's immune system directly to a human's without the math getting in the way.

2. No Coding Required (The Web App)
The authors didn't just write a math paper; they built a website (like a calculator for biology).

• You can upload your data, click a few buttons, and get a graph showing if a patient is likely to survive or if a drug is working. You don't need to know how to code; you just need to know your biology.

3. Predicting the Future (Survival Analysis)
Because the scores are so clear, COMPASS can predict outcomes.

• In the paper, they used it on sepsis (a life-threatening infection) patients. The COMPASS score acted like a crystal ball: Patients with a high score had a much higher risk of death. It turned a complex gene list into a simple "High Risk / Low Risk" warning system.

The Bottom Line

COMPASS takes the chaotic noise of gene expression and turns it into a clear, deterministic signal.

• Old Way: "This group looks different from that group, maybe?" (Relative, shaky, hard to repeat).
• COMPASS Way: "Gene X is definitely ON, Gene Y is definitely OFF, and the total score is 85." (Absolute, stable, repeatable).

It bridges the gap between the lab bench and the hospital bed, allowing doctors and researchers to use the same "language" to understand disease, test drugs, and save lives.

Technical Summary

1. Problem Statement

Current methods for quantifying pathway activity in systems biology and precision medicine face three primary limitations:

• Reliance on Permutation and Ontologies: Widely used enrichment methods (e.g., GSEA, GSVA, ssGSEA) rely on ranked gene lists, random permutations, and evolving biological ontologies. This introduces variability across datasets, platforms, and time, limiting reproducibility.
• Loss of Directionality: Many clinically relevant gene signatures contain both upregulated and downregulated genes. Treating these as undirected sets often conflates opposing biological signals, leading to inconsistent interpretations.
• Lack of Accessibility and Integration: Existing tools often require coding expertise (R/Python) or complex statistical modeling (e.g., matrix factorization, deep learning), creating barriers for non-computational users. Furthermore, they often fail to seamlessly link molecular measurements to clinical outcomes like survival.

2. Methodology: The COMPASS Framework

The authors introduce COMPASS (COMPosite Activity Scoring System), a deterministic, ontology-free, and threshold-based framework. It converts raw gene expression data into per-sample pathway activity scores using a closed-form, three-step mathematical process:

1. Threshold Determination (StepMiner Algorithm):

   • For every gene, COMPASS identifies an intrinsic expression boundary (transition index, $T_g$) where expression transitions from "off" (low) to "on" (high).
   • This is achieved by fitting a step function to ordered expression values to minimize within-group variance.
   • A conservative confidence offset ($\sigma$) is applied to define a final decision boundary ($T^*_g = T_g + \sigma$), ensuring only confident deviations contribute to the score.

2. Standardization:

   • Gene expression deviations from the threshold are standardized relative to the gene-specific standard deviation ($\sigma_g$).
   • The formula used is $Z_{g,i} = (E_{g,i} - T^*_g) / (3\sigma_g)$.
   • Dividing by $3\sigma_g$ compresses extreme outliers and brings all genes to a comparable dynamic scale while preserving biological directionality.

3. Composite Aggregation:

   • For a defined gene set $G$, the composite activity score ($C_i$) for sample $i$ is calculated as the weighted mean of standardized deviations:
     $$C_i = \frac{1}{m} \sum_{j=1}^{m} w_j Z_{j,i}$$
   • Directionality: Weights ($w_j$) are assigned as $+1$ for activation genes and $-1$ for repression genes. This allows the integration of bidirectional gene programs into a single net score.
   • The result is a continuous, direction-aware score where positive values indicate pathway activation and negative values indicate repression.

Implementation:

• Platform: Implemented as a web-based application (https://compass.precsn.com/) requiring no coding.
• Inputs: Users upload gene expression matrices (TPM/CPM), define gene signatures with directionality, and optionally provide clinical metadata (e.g., survival time).
• Outputs: Automated generation of boxplots, ROC-AUC analyses, and survival curves (Kaplan-Meier, Cox proportional hazards).

3. Key Contributions

• Deterministic & Reproducible: Unlike permutation-based methods, COMPASS produces identical outputs for identical inputs, ensuring auditability and reproducibility across runs and environments.
• Ontology-Free: It does not rely on predefined pathway databases; users can define custom gene sets, making it adaptable to novel biological discoveries.
• Bidirectional Integration: It uniquely handles opposing gene signals (up/down) within a single signature, preventing signal cancellation or conflation.
• Clinical Translation: It bridges the gap between molecular data and clinical outcomes by generating "digital biomarkers" that can be directly used for survival analysis and risk stratification without intermediate model fitting.
• Accessibility: The web interface democratizes advanced pathway analysis for non-computational biologists and clinicians.

4. Results

The authors validated COMPASS across diverse datasets and compared it against GSVA and ssGSEA:

• Cross-Model Benchmarking: COMPASS successfully quantified pathway activity across human cohorts, animal models (mice/hamsters), and organoids.
  • Example: In colorectal cancer, a 50-gene stemness signature consistently detected therapeutic efficacy across cell lines, xenografts, and patient-derived organoids.
  • Example: In respiratory viral injury, a lung damage signature showed conserved induction across human biopsies, hamster lungs, and lung organoids.
• Head-to-Head Comparison (Sepsis Cohorts):
  • Using an 11-gene sepsis signature across 10 independent cohorts (531 samples), COMPASS demonstrated superior and consistent discrimination compared to GSVA and ssGSEA.
  • Robustness: Stratified bootstrap resampling ($n=1000$) showed COMPASS had tighter AUC distributions and higher central tendency.
  • Performance Metrics: COMPASS achieved a mean specificity of 0.92 and sensitivity of 0.91, closely matching the performance of the FDA-approved SEPSIS-SHIELD clinical study. In contrast, GSVA (0.69/0.69) and ssGSEA (0.80/0.88) showed lower and less balanced performance, particularly when analyzing directional subsets of genes.
• Outcome Prediction:
  • Using a 12-gene sepsis severity signature on a multi-cohort dataset (GSE310929, $n=593$), COMPASS scores stratified patients into high-risk and low-risk groups.
  • High scores correlated with significantly worse survival (Hazard Ratio = 2.32), demonstrating the utility of COMPASS scores for time-to-event analysis.

5. Significance

• Paradigm Shift: COMPASS moves pathway analysis from "relative enrichment" (ranking genes within a cohort) to "absolute activity" (measuring deviation from intrinsic biological thresholds).
• Regulatory & Translational Relevance: By providing stable, reproducible, and interpretable digital biomarkers, COMPASS supports the use of New Approach Methodologies (NAMs) like organoids and animal models for regulatory decision-making by anchoring them to human clinical data.
• Unified Framework: It unifies classification, cross-cohort benchmarking, and survival modeling into a single deterministic system, eliminating the need for stochastic inference.
• Generalizability: The framework is applicable to oncology, infectious disease, and inflammatory disorders, offering a scalable solution for precision medicine that bridges the gap between mechanistic biology and clinical outcomes.

In conclusion, COMPASS offers a transparent, mathematically rigorous, and accessible alternative to traditional enrichment methods, enabling the generation of robust digital biomarkers that are directly actionable in clinical and translational settings.